From f127b67e731571d4e8c7574be3ef0451582daa62 Mon Sep 17 00:00:00 2001
From: MMEXA <youlancelot@gmail.com>
Date: Sun, 12 Jul 2026 20:34:31 +0800
Subject: [PATCH 01/12] fix(openai): encode tool errors in Responses output

---
 .../src/formats/conversion/request.rs         | 68 ++++++++++++++++++
 .../src/formats/openai/responses/mod.rs       | 20 ++++++
 .../src/formats/openai/responses/request.rs   | 71 +++++++++++++++----
 .../src/formats/openai/responses/response.rs  | 42 +++++++++--
 4 files changed, 182 insertions(+), 19 deletions(-)

diff --git a/crates/aether-ai-formats/src/formats/conversion/request.rs b/crates/aether-ai-formats/src/formats/conversion/request.rs
index f339c27dc..993cd67db 100644
--- a/crates/aether-ai-formats/src/formats/conversion/request.rs
+++ b/crates/aether-ai-formats/src/formats/conversion/request.rs
@@ -1024,6 +1024,74 @@ mod tests {
         );
     }
 
+    #[test]
+    fn claude_request_to_responses_encodes_error_tool_results_in_output() {
+        let body = json!({
+            "model": "claude-sonnet",
+            "messages": [{
+                "role": "user",
+                "content": [
+                    {
+                        "type": "tool_result",
+                        "tool_use_id": "toolu_error_string",
+                        "content": "lookup failed",
+                        "is_error": true
+                    },
+                    {
+                        "type": "tool_result",
+                        "tool_use_id": "toolu_error_empty",
+                        "content": "",
+                        "is_error": true
+                    },
+                    {
+                        "type": "tool_result",
+                        "tool_use_id": "toolu_error_object",
+                        "content": {"code": "ENOENT"},
+                        "is_error": true
+                    },
+                    {
+                        "type": "tool_result",
+                        "tool_use_id": "toolu_error_image",
+                        "content": [
+                            {"type": "text", "text": "preview failed"},
+                            {
+                                "type": "image",
+                                "source": {
+                                    "type": "base64",
+                                    "media_type": "image/png",
+                                    "data": "AAAA"
+                                }
+                            }
+                        ],
+                        "is_error": true
+                    }
+                ]
+            }],
+            "max_tokens": 128,
+        });
+
+        let converted = registry::convert_request(
+            "claude:messages",
+            "openai:responses",
+            &body,
+            &FormatContext::default(),
+        )
+        .expect("responses request");
+        let input = converted["input"].as_array().expect("responses input");
+
+        assert_eq!(input[0]["output"], "[tool error]\nlookup failed");
+        assert_eq!(input[1]["output"], "[tool error]");
+        assert_eq!(input[2]["output"], "[tool error]\n{\"code\":\"ENOENT\"}");
+        assert_eq!(input[3]["output"], "[tool error]\npreview failed");
+        assert_eq!(input[4]["role"], "user");
+        assert_eq!(input[4]["content"][0]["type"], "input_image");
+        assert_eq!(
+            input[4]["content"][0]["image_url"],
+            "data:image/png;base64,AAAA"
+        );
+        assert!(input.iter().all(|item| item.get("is_error").is_none()));
+    }
+
     #[test]
     fn claude_request_to_responses_rejects_unrepresentable_tool_result_blocks() {
         let body = json!({
diff --git a/crates/aether-ai-formats/src/formats/openai/responses/mod.rs b/crates/aether-ai-formats/src/formats/openai/responses/mod.rs
index 1c9d73917..16e73a0c7 100644
--- a/crates/aether-ai-formats/src/formats/openai/responses/mod.rs
+++ b/crates/aether-ai-formats/src/formats/openai/responses/mod.rs
@@ -1,5 +1,25 @@
+use serde_json::Value;
+
 pub mod codex;
 pub mod request;
 pub mod response;
 pub mod spec;
 pub mod stream;
+
+const TOOL_ERROR_PREFIX: &str = "[tool error]";
+
+fn encode_tool_result_error(output: Value, is_error: bool) -> Value {
+    if !is_error {
+        return output;
+    }
+    let detail = match output {
+        Value::String(text) => text,
+        Value::Null => String::new(),
+        value => serde_json::to_string(&value).unwrap_or_else(|_| value.to_string()),
+    };
+    if detail.is_empty() {
+        Value::String(TOOL_ERROR_PREFIX.to_string())
+    } else {
+        Value::String(format!("{TOOL_ERROR_PREFIX}\n{detail}"))
+    }
+}
diff --git a/crates/aether-ai-formats/src/formats/openai/responses/request.rs b/crates/aether-ai-formats/src/formats/openai/responses/request.rs
index ef70667fc..519a2e505 100644
--- a/crates/aether-ai-formats/src/formats/openai/responses/request.rs
+++ b/crates/aether-ai-formats/src/formats/openai/responses/request.rs
@@ -2,6 +2,8 @@ use std::collections::{BTreeMap, VecDeque};
 
 use serde_json::{json, Map, Value};
 
+use super::encode_tool_result_error;
+
 use crate::{
     formats::context::FormatContext,
     formats::openai::shared::map_thinking_budget_to_openai_reasoning_effort,
@@ -449,6 +451,7 @@ fn canonical_messages_to_responses_input(canonical: &CanonicalRequest) -> Option
                     let (tool_output, extra_user_content) = responses_tool_result_payload(
                         output.as_ref(),
                         content_text.as_deref(),
+                        *is_error,
                         extensions,
                     )?;
                     let call_id =
@@ -464,9 +467,6 @@ fn canonical_messages_to_responses_input(canonical: &CanonicalRequest) -> Option
                     );
                     item.insert("call_id".to_string(), Value::String(call_id));
                     item.insert("output".to_string(), tool_output);
-                    if *is_error {
-                        item.insert("is_error".to_string(), Value::Bool(true));
-                    }
                     let extension_fields =
                         openai_responses_item_extension_object(extensions, &item);
                     item.extend(extension_fields);
@@ -1133,18 +1133,19 @@ fn canonical_tool_choice_to_responses(
 fn responses_tool_result_payload(
     output: Option<&Value>,
     content_text: Option<&str>,
+    is_error: bool,
     extensions: &BTreeMap<String, Value>,
 ) -> Option<(Value, Vec<Value>)> {
     if let Some(Value::Array(parts)) = output {
         if is_claude_tool_result(extensions) {
-            return claude_tool_result_parts_to_responses_payload(parts);
+            return claude_tool_result_parts_to_responses_payload(parts, is_error);
         }
         if let Some(output) = openai_chat_tool_result_parts_to_responses_output(parts) {
-            return Some((output, Vec::new()));
+            return Some((encode_tool_result_error(output, is_error), Vec::new()));
         }
     }
     Some((
-        responses_tool_result_output(output, content_text),
+        responses_tool_result_output(output, content_text, is_error),
         Vec::new(),
     ))
 }
@@ -1260,14 +1261,22 @@ fn openai_chat_tool_result_fallback_part(part: &Value) -> Value {
     })
 }
 
-fn responses_tool_result_output(output: Option<&Value>, content_text: Option<&str>) -> Value {
+fn responses_tool_result_output(
+    output: Option<&Value>,
+    content_text: Option<&str>,
+    is_error: bool,
+) -> Value {
     let text = match output {
         Some(Value::String(text)) => text.clone(),
         Some(Value::Null) => String::new(),
         Some(value) => serde_json::to_string(value).unwrap_or_default(),
         None => content_text.unwrap_or_default().to_string(),
     };
-    Value::String(non_empty_responses_tool_output(&text))
+    let output = encode_tool_result_error(Value::String(text), is_error);
+    match output {
+        Value::String(text) => Value::String(non_empty_responses_tool_output(&text)),
+        output => output,
+    }
 }
 
 pub(crate) fn claude_tool_result_parts_are_openai_responses_representable(parts: &[Value]) -> bool {
@@ -1322,7 +1331,10 @@ fn claude_document_block_is_openai_responses_representable(block: &Map<String, V
     }
 }
 
-fn claude_tool_result_parts_to_responses_payload(parts: &[Value]) -> Option<(Value, Vec<Value>)> {
+fn claude_tool_result_parts_to_responses_payload(
+    parts: &[Value],
+    is_error: bool,
+) -> Option<(Value, Vec<Value>)> {
     let mut output_texts = Vec::new();
     let mut extra_user_content = Vec::new();
 
@@ -1365,10 +1377,12 @@ fn claude_tool_result_parts_to_responses_payload(parts: &[Value]) -> Option<(Val
         }
     }
 
-    Some((
-        Value::String(non_empty_responses_tool_output(&output_texts.join("\n\n"))),
-        extra_user_content,
-    ))
+    let output = encode_tool_result_error(Value::String(output_texts.join("\n\n")), is_error);
+    let output = match output {
+        Value::String(text) => Value::String(non_empty_responses_tool_output(&text)),
+        output => output,
+    };
+    Some((output, extra_user_content))
 }
 
 fn claude_image_block_to_responses_input_part(block: &Map<String, Value>) -> Option<Value> {
@@ -1661,6 +1675,37 @@ mod tests {
         assert_eq!(body["input"][0]["output"], "(empty)");
     }
 
+    #[test]
+    fn responses_request_encodes_tool_errors_for_regular_and_compact_requests() {
+        let request = CanonicalRequest {
+            model: "gpt-5.6-sol".to_string(),
+            messages: vec![CanonicalMessage {
+                role: CanonicalRole::Tool,
+                content: vec![CanonicalContentBlock::ToolResult {
+                    tool_use_id: "call_error".to_string(),
+                    name: None,
+                    output: Some(json!("command failed")),
+                    content_text: None,
+                    is_error: true,
+                    extensions: Default::default(),
+                }],
+                extensions: Default::default(),
+            }],
+            ..CanonicalRequest::default()
+        };
+
+        for compact in [false, true] {
+            let body =
+                to_raw(&request, "gpt-5.6-sol", false, compact).expect("Responses request body");
+            let item = &body["input"][0];
+
+            assert_eq!(item["type"], "function_call_output");
+            assert_eq!(item["call_id"], "call_error");
+            assert_eq!(item["output"], "[tool error]\ncommand failed");
+            assert!(item.get("is_error").is_none());
+        }
+    }
+
     #[test]
     fn responses_request_replaces_empty_tool_call_identifiers() {
         let request = CanonicalRequest {
diff --git a/crates/aether-ai-formats/src/formats/openai/responses/response.rs b/crates/aether-ai-formats/src/formats/openai/responses/response.rs
index 12048cda0..088ccc60d 100644
--- a/crates/aether-ai-formats/src/formats/openai/responses/response.rs
+++ b/crates/aether-ai-formats/src/formats/openai/responses/response.rs
@@ -5,6 +5,8 @@ use std::{
 
 use serde_json::{json, Map, Value};
 
+use super::encode_tool_result_error;
+
 use crate::{
     formats::context::FormatContext,
     protocol::canonical::{
@@ -270,13 +272,13 @@ pub fn to_raw(canonical: &CanonicalResponse, report_context: &Value, compact: bo
                 item.insert("call_id".to_string(), Value::String(tool_use_id.clone()));
                 item.insert(
                     "output".to_string(),
-                    result_output
-                        .clone()
-                        .unwrap_or_else(|| Value::String(content_text.clone().unwrap_or_default())),
+                    encode_tool_result_error(
+                        result_output.clone().unwrap_or_else(|| {
+                            Value::String(content_text.clone().unwrap_or_default())
+                        }),
+                        *is_error,
+                    ),
                 );
-                if *is_error {
-                    item.insert("is_error".to_string(), Value::Bool(true));
-                }
                 let extension_fields = openai_responses_item_extension_object(extensions, &item);
                 item.extend(extension_fields);
                 output.push(Value::Object(item));
@@ -618,6 +620,34 @@ mod tests {
         assert_eq!(body["conversation"]["id"], "conv_123");
     }
 
+    #[test]
+    fn responses_response_builder_encodes_tool_errors_in_output() {
+        let response = CanonicalResponse {
+            id: "resp_tool_error".to_string(),
+            model: "gpt-5.6-sol".to_string(),
+            content: vec![CanonicalContentBlock::ToolResult {
+                tool_use_id: "call_error".to_string(),
+                name: None,
+                output: Some(json!("command failed")),
+                content_text: None,
+                is_error: true,
+                extensions: BTreeMap::new(),
+            }],
+            outputs: Vec::new(),
+            stop_reason: Some(CanonicalStopReason::EndTurn),
+            usage: None,
+            extensions: BTreeMap::new(),
+        };
+
+        let body = to_raw(&response, &json!({}), false);
+        let item = &body["output"][0];
+
+        assert_eq!(item["type"], "function_call_output");
+        assert_eq!(item["call_id"], "call_error");
+        assert_eq!(item["output"], "[tool error]\ncommand failed");
+        assert!(item.get("is_error").is_none());
+    }
+
     #[test]
     fn compact_response_builder_emits_the_compaction_resource_shape() {
         let mut extensions = BTreeMap::new();

From b1be370b2e9a34a38b5d0f41b6ca8b0c79fdf1b7 Mon Sep 17 00:00:00 2001
From: MMEXA <youlancelot@gmail.com>
Date: Sun, 12 Jul 2026 21:08:30 +0800
Subject: [PATCH 02/12] fix(gateway): scope concurrency test helper to tests

---
 apps/aether-gateway/src/main.rs | 1 +
 1 file changed, 1 insertion(+)

diff --git a/apps/aether-gateway/src/main.rs b/apps/aether-gateway/src/main.rs
index 68d1413f3..00e15c825 100644
--- a/apps/aether-gateway/src/main.rs
+++ b/apps/aether-gateway/src/main.rs
@@ -286,6 +286,7 @@ fn available_parallelism_usize() -> usize {
         .max(1)
 }
 
+#[cfg(test)]
 fn automatic_gateway_request_concurrency_for_parallelism(parallelism: usize) -> usize {
     automatic_gateway_request_concurrency_for_capacity(parallelism, None)
 }

From fc2dfb82d2aad9bf5454ad16c22ddb16d2614087 Mon Sep 17 00:00:00 2001
From: MMEXA <youlancelot@gmail.com>
Date: Sun, 12 Jul 2026 23:04:33 +0800
Subject: [PATCH 03/12] fix(codex): preserve reset consume request body

---
 .../src/control/tests/admin_endpoints.rs      | 19 +++++++++++++++++++
 .../src/handlers/shared/request_utils.rs      |  5 +++++
 2 files changed, 24 insertions(+)

diff --git a/apps/aether-gateway/src/control/tests/admin_endpoints.rs b/apps/aether-gateway/src/control/tests/admin_endpoints.rs
index cc43cdb94..9594c7221 100644
--- a/apps/aether-gateway/src/control/tests/admin_endpoints.rs
+++ b/apps/aether-gateway/src/control/tests/admin_endpoints.rs
@@ -454,6 +454,25 @@ fn classifies_admin_codex_reset_credit_consume_as_admin_proxy_route() {
     assert!(!decision.is_execution_runtime_candidate());
 }
 
+#[test]
+fn admin_codex_reset_credit_consume_buffers_idempotency_key_body() {
+    let headers = headers(&[]);
+    let uri: Uri = "/api/admin/endpoints/keys/key-codex/codex-reset-credit/consume"
+        .parse()
+        .expect("uri should parse");
+    let decision = classify_control_route(&http::Method::POST, &uri, &headers)
+        .expect("decision should resolve");
+    let context = GatewayPublicRequestContext::from_request_parts(
+        "trace-codex-reset-credit-consume",
+        &http::Method::POST,
+        &uri,
+        &headers,
+        Some(decision),
+    );
+
+    assert!(local_proxy_route_requires_buffered_body(&context));
+}
+
 #[test]
 fn admin_refresh_provider_quota_buffers_request_body_for_key_selection() {
     let headers = headers(&[]);
diff --git a/apps/aether-gateway/src/handlers/shared/request_utils.rs b/apps/aether-gateway/src/handlers/shared/request_utils.rs
index aba7e9606..9cf01b71b 100644
--- a/apps/aether-gateway/src/handlers/shared/request_utils.rs
+++ b/apps/aether-gateway/src/handlers/shared/request_utils.rs
@@ -221,6 +221,11 @@ pub(crate) fn admin_proxy_local_requires_buffered_body(
                 | (Some("endpoints_manage"), http::Method::POST, Some("create_endpoint"))
                 | (Some("endpoints_manage"), http::Method::POST, Some("batch_delete_keys"))
                 | (Some("endpoints_manage"), http::Method::POST, Some("refresh_quota"))
+                | (
+                    Some("endpoints_manage"),
+                    http::Method::POST,
+                    Some("codex_reset_credit_consume"),
+                )
                 | (Some("endpoints_manage"), http::Method::PUT, Some("update_key"))
                 | (Some("endpoints_manage"), http::Method::PUT, Some("update_endpoint"))
                 | (Some("modules_manage"), http::Method::PUT, Some("set_enabled"))

From e8afa03e454b59890a166e45e5eaf2a8ed02b5cc Mon Sep 17 00:00:00 2001
From: MMEXA <youlancelot@gmail.com>
Date: Mon, 13 Jul 2026 06:07:54 +0800
Subject: [PATCH 04/12] =?UTF-8?q?fix(auth):=20=E6=8E=88=E6=9D=83=20Respons?=
 =?UTF-8?q?es=20Compact=20=E4=BC=B4=E9=9A=8F=E7=AB=AF=E7=82=B9?=
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

---
 crates/aether-ai-formats/src/formats/id.rs    | 30 +++++++++++++++++--
 .../aether-data/src/repository/auth/types.rs  | 16 +++++++++-
 2 files changed, 43 insertions(+), 3 deletions(-)

diff --git a/crates/aether-ai-formats/src/formats/id.rs b/crates/aether-ai-formats/src/formats/id.rs
index fc83ffdc4..c45c8f66e 100644
--- a/crates/aether-ai-formats/src/formats/id.rs
+++ b/crates/aether-ai-formats/src/formats/id.rs
@@ -170,7 +170,11 @@ pub fn api_format_permission_covers(allowed_value: &str, requested_api_format: &
     !allowed_value.is_empty()
         && !requested_api_format.is_empty()
         && (allowed_value == requested_api_format
-            || allowed_value == "openai:responses" && requested_api_format == "openai:search")
+            || allowed_value == "openai:responses"
+                && matches!(
+                    requested_api_format.as_str(),
+                    "openai:responses:compact" | "openai:search"
+                ))
 }
 
 pub fn intersect_api_format_allowed_lists(left: &[String], right: &[String]) -> Vec<String> {
@@ -269,11 +273,15 @@ mod tests {
     }
 
     #[test]
-    fn responses_permission_covers_only_its_search_companion() {
+    fn responses_permission_covers_its_companion_endpoints() {
         assert!(api_format_permission_covers(
             "OPENAI:RESPONSES",
             "openai:search"
         ));
+        assert!(api_format_permission_covers(
+            "OPENAI:RESPONSES",
+            "openai:responses:compact"
+        ));
         assert!(api_format_permission_covers(
             "openai:search",
             "openai:search"
@@ -282,6 +290,10 @@ mod tests {
             "openai:search",
             "openai:responses"
         ));
+        assert!(!api_format_permission_covers(
+            "openai:responses:compact",
+            "openai:responses"
+        ));
         assert!(!api_format_permission_covers(
             "openai:responses",
             "openai:chat"
@@ -290,6 +302,13 @@ mod tests {
             api_format_permission_storage_aliases("openai:search"),
             vec!["openai:search".to_string(), "openai:responses".to_string()]
         );
+        assert_eq!(
+            api_format_permission_storage_aliases("openai:responses:compact"),
+            vec![
+                "openai:responses:compact".to_string(),
+                "openai:responses".to_string(),
+            ]
+        );
         assert_eq!(
             api_format_permission_storage_aliases("openai:responses"),
             vec!["openai:responses".to_string()]
@@ -350,6 +369,13 @@ mod tests {
             ),
             vec!["openai:search".to_string()]
         );
+        assert_eq!(
+            intersect_api_format_allowed_lists(
+                &["openai:responses".to_string()],
+                &["openai:responses:compact".to_string()],
+            ),
+            vec!["openai:responses:compact".to_string()]
+        );
         assert_eq!(
             intersect_api_format_allowed_lists(
                 &["openai:search".to_string()],
diff --git a/crates/aether-data/src/repository/auth/types.rs b/crates/aether-data/src/repository/auth/types.rs
index f3b7039b4..899e2c4fe 100644
--- a/crates/aether-data/src/repository/auth/types.rs
+++ b/crates/aether-data/src/repository/auth/types.rs
@@ -827,7 +827,7 @@ mod tests {
     };
 
     #[test]
-    fn api_format_policy_intersection_preserves_search_companion_scope() {
+    fn api_format_policy_intersection_preserves_companion_scope() {
         assert_eq!(
             aether_ai_formats::intersect_api_format_allowed_lists(
                 &["openai:search".to_string()],
@@ -842,6 +842,20 @@ mod tests {
             ),
             vec!["openai:search".to_string()]
         );
+        assert_eq!(
+            aether_ai_formats::intersect_api_format_allowed_lists(
+                &["openai:responses".to_string()],
+                &["openai:responses:compact".to_string()],
+            ),
+            vec!["openai:responses:compact".to_string()]
+        );
+        assert_eq!(
+            aether_ai_formats::intersect_api_format_allowed_lists(
+                &["openai:responses:compact".to_string()],
+                &["openai:responses".to_string()],
+            ),
+            vec!["openai:responses:compact".to_string()]
+        );
         assert!(aether_ai_formats::intersect_api_format_allowed_lists(
             &["openai:search".to_string()],
             &["openai:chat".to_string()],

From 2fc604e047a63103232e0066780f24556c0bb36f Mon Sep 17 00:00:00 2001
From: MMEXA <youlancelot@gmail.com>
Date: Mon, 13 Jul 2026 21:43:18 +0800
Subject: [PATCH 05/12] =?UTF-8?q?feat(codex):=20=E6=8C=89=E6=93=8D?=
 =?UTF-8?q?=E4=BD=9C=E8=AF=AD=E4=B9=89=E8=B7=AF=E7=94=B1=20Responses=20V2?=
 =?UTF-8?q?=20=E5=8E=8B=E7=BC=A9?=
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

---
 .../planner/candidate_materialization.rs      |   6 +
 .../ai_serving/planner/candidate_source.rs    | 112 ++++++++++--
 .../planner/standard/family/candidates.rs     |   3 +
 .../standard/openai/chat/decision/support.rs  |   1 +
 .../standard/openai/chat/plans/candidates.rs  |   1 +
 .../openai/responses/decision/support.rs      |   9 +-
 .../src/ai_serving/planner/state/scheduler.rs |  35 +++-
 .../aether-gateway/src/ai_serving/pure/mod.rs |   2 +-
 .../src/cache/candidate_page.rs               |  31 ++++
 apps/aether-gateway/src/control/auth/gate.rs  |   2 +
 .../src/data/candidate_selection.rs           |  24 +++
 apps/aether-gateway/src/data/tests.rs         |   2 +
 .../query/models/model_test/model_mapping.rs  |  18 ++
 .../src/handlers/admin/request/models.rs      |   9 +-
 .../handlers/public/support/user_me_usage.rs  |   2 +
 .../src/scheduler/candidate/enumeration.rs    |   6 +-
 .../src/scheduler/candidate/mod.rs            |  37 ++++
 .../src/scheduler/candidate/selection.rs      |   4 +
 .../src/scheduler/candidate/tests/affinity.rs |   3 +
 .../src/scheduler/candidate/tests/model.rs    |   6 +
 .../scheduler/candidate/tests/selection.rs    |   5 +
 .../src/scheduler/candidate/tests/support.rs  |   2 +
 apps/aether-gateway/src/state/catalog.rs      |   1 +
 apps/aether-gateway/src/testkit.rs            |   1 +
 .../src/tests/ai_execute/finalize_local.rs    |   6 +
 .../ai_execute/finalize_local_cli/compact.rs  |   1 +
 .../finalize_local_cli/cross_format.rs        |   3 +
 .../ai_execute/finalize_local_cli/direct.rs   |   2 +
 .../finalize_local_provider/claude.rs         |   3 +
 .../finalize_local_provider/gemini.rs         |   4 +
 .../src/tests/ai_execute/lifecycle.rs         |   1 +
 .../src/tests/ai_execute/stream/decision.rs   |   5 +
 .../src/tests/ai_execute/stream/image.rs      |   3 +
 .../tests/ai_execute/stream/pii_redaction.rs  |   1 +
 .../tests/ai_execute/stream_cli/compact.rs    |   1 +
 .../src/tests/ai_execute/stream_cli/direct.rs |   1 +
 .../src/tests/ai_execute/stream_provider.rs   |   4 +
 .../stream_provider_gemini/local_chat.rs      |   1 +
 .../stream_provider_gemini/local_cli.rs       |   4 +
 .../tests/ai_execute/sync/chat/failover.rs    |   4 +
 .../ai_execute/sync/chat/local_decision.rs    |  11 ++
 .../ai_execute/sync/chat/pii_redaction.rs     |   1 +
 .../ai_execute/sync/claude/claude_code.rs     |   1 +
 .../src/tests/ai_execute/sync/claude/kiro.rs  |   2 +
 .../ai_execute/sync/claude/local_chat.rs      |   2 +
 .../tests/ai_execute/sync/claude/local_cli.rs |   3 +
 .../src/tests/ai_execute/sync/cli.rs          |   9 +
 .../src/tests/ai_execute/sync/gemini/cli.rs   |   5 +
 .../ai_execute/sync/gemini/local_chat.rs      |   2 +
 .../src/tests/ai_execute/sync/image.rs        |   4 +
 .../ai_execute/sync/pii_redaction_formats.rs  |   1 +
 .../src/tests/ai_execute/sync/search.rs       |   1 +
 apps/aether-gateway/src/tests/audit.rs        |   1 +
 apps/aether-gateway/src/tests/files/mod.rs    |   1 +
 apps/aether-gateway/src/tests/frontdoor.rs    |   1 +
 apps/aether-gateway/src/tests/frontdoor/ai.rs |   1 +
 apps/aether-gateway/src/tests/usage.rs        |   1 +
 apps/aether-gateway/src/tests/usage/local.rs  |   2 +
 .../aether-gateway/src/tests/usage/pricing.rs |   1 +
 .../src/tests/video/gemini_sync_create.rs     |   1 +
 .../src/tests/video/openai_sync_create.rs     |   1 +
 .../aether-admin/src/observability/usage.rs   |   2 +
 .../src/formats/openai/responses/mod.rs       |  66 +++++++
 crates/aether-ai-formats/src/lib.rs           |   3 +
 .../repository/candidate_selection/types.rs   |   4 +
 .../repository/candidate_selection/memory.rs  |   3 +
 .../repository/candidate_selection/mysql.rs   |  17 ++
 .../candidate_selection/postgres.rs           |  23 ++-
 .../repository/candidate_selection/sqlite.rs  |  27 ++-
 .../src/candidate/enumeration.rs              |   4 +-
 .../src/candidate/mod.rs                      |   5 +
 .../src/candidate/types.rs                    |   1 +
 crates/aether-scheduler-core/src/lib.rs       |  11 +-
 crates/aether-scheduler-core/src/model.rs     | 169 +++++++++++++++---
 crates/aether-usage-runtime/src/write.rs      |  29 ++-
 75 files changed, 732 insertions(+), 50 deletions(-)

diff --git a/apps/aether-gateway/src/ai_serving/planner/candidate_materialization.rs b/apps/aether-gateway/src/ai_serving/planner/candidate_materialization.rs
index 64fce324d..1648ada25 100644
--- a/apps/aether-gateway/src/ai_serving/planner/candidate_materialization.rs
+++ b/apps/aether-gateway/src/ai_serving/planner/candidate_materialization.rs
@@ -706,6 +706,7 @@ pub(crate) async fn build_lazy_requested_model_execution_candidate_attempt_sourc
     trace_id: &str,
     client_api_format: &str,
     requested_model: &str,
+    request_operation: Option<&str>,
     require_streaming: bool,
     auth_snapshot: &GatewayAuthApiKeySnapshot,
     client_session_affinity: Option<&ClientSessionAffinity>,
@@ -734,6 +735,7 @@ where
         model_directive_policy,
         client_api_format,
         requested_model,
+        request_operation,
         require_streaming,
         required_capabilities,
         auth_snapshot,
@@ -1141,6 +1143,7 @@ async fn resolve_priority_candidate_page_with_cache(
 
     let key = CandidateResolvedPageCacheKey::new(
         &cursor.requested_model,
+        cursor.page_cursor.resolved_page_cache_request_operation(),
         &cursor.client_api_format,
         true,
         &cursor.auth_snapshot,
@@ -2190,6 +2193,7 @@ mod tests {
             &model_directive_policy,
             "openai:chat",
             "gpt-5",
+            None,
             true,
             None,
             &auth_snapshot,
@@ -2245,6 +2249,7 @@ mod tests {
             &model_directive_policy,
             "openai:chat",
             "gpt-5",
+            None,
             true,
             None,
             &auth_snapshot,
@@ -2282,6 +2287,7 @@ mod tests {
             &model_directive_policy,
             "openai:chat",
             "gpt-5",
+            None,
             true,
             None,
             &auth_snapshot,
diff --git a/apps/aether-gateway/src/ai_serving/planner/candidate_source.rs b/apps/aether-gateway/src/ai_serving/planner/candidate_source.rs
index 1ff3fc292..ba48d33eb 100644
--- a/apps/aether-gateway/src/ai_serving/planner/candidate_source.rs
+++ b/apps/aether-gateway/src/ai_serving/planner/candidate_source.rs
@@ -6,9 +6,10 @@ use aether_routing_core::ResolvedRoutingPolicy;
 use aether_runtime::ConcurrencyPermit;
 use aether_scheduler_core::{
     enumerate_minimal_candidate_selection_with_model_directives, normalize_api_format,
-    resolve_requested_global_model_name_with_model_directives,
-    row_supports_requested_model_with_model_directives, ClientSessionAffinity,
-    EnumerateMinimalCandidateSelectionInput, SchedulerMinimalCandidateSelectionCandidate,
+    resolve_requested_global_model_name_with_model_directives_and_request_operation,
+    row_supports_requested_model_with_model_directives_and_request_operation,
+    ClientSessionAffinity, EnumerateMinimalCandidateSelectionInput,
+    SchedulerMinimalCandidateSelectionCandidate,
 };
 use async_trait::async_trait;
 use std::collections::{BTreeMap, BTreeSet, VecDeque};
@@ -56,6 +57,7 @@ struct GatewayLocalCandidatePreselectionPort<'a> {
     state: PlannerAppState<'a>,
     client_api_format: &'a str,
     requested_model: &'a str,
+    request_operation: Option<&'a str>,
     require_streaming: bool,
     required_capabilities: Option<&'a serde_json::Value>,
     auth_snapshot: &'a GatewayAuthApiKeySnapshot,
@@ -112,7 +114,7 @@ impl AiCandidatePreselectionPort for GatewayLocalCandidatePreselectionPort<'_> {
         let auth_snapshot = matches_client_format.then_some(self.auth_snapshot);
         let (candidates, skipped_candidates) = self
             .state
-            .list_selectable_candidates_with_skip_reasons(
+            .list_selectable_candidates_with_skip_reasons_for_request_operation(
                 candidate_api_format,
                 self.routing_model(candidate_api_format),
                 self.require_streaming,
@@ -121,6 +123,7 @@ impl AiCandidatePreselectionPort for GatewayLocalCandidatePreselectionPort<'_> {
                 self.client_session_affinity,
                 self.ranking_seed,
                 false,
+                self.request_operation,
             )
             .await?;
 
@@ -197,6 +200,7 @@ pub(crate) async fn preselect_local_execution_candidates_with_serving(
     model_directive_policy: &crate::system_features::ModelDirectivePolicySnapshot,
     client_api_format: &str,
     requested_model: &str,
+    request_operation: Option<&str>,
     require_streaming: bool,
     required_capabilities: Option<&serde_json::Value>,
     auth_snapshot: &GatewayAuthApiKeySnapshot,
@@ -221,6 +225,7 @@ pub(crate) async fn preselect_local_execution_candidates_with_serving(
         model_directive_policy,
         client_api_format,
         requested_model,
+        request_operation,
         require_streaming,
         required_capabilities,
         auth_snapshot,
@@ -239,6 +244,7 @@ pub(crate) async fn preselect_local_execution_candidates_for_api_formats_with_se
     model_directive_policy: &crate::system_features::ModelDirectivePolicySnapshot,
     client_api_format: &str,
     requested_model: &str,
+    request_operation: Option<&str>,
     require_streaming: bool,
     required_capabilities: Option<&serde_json::Value>,
     auth_snapshot: &GatewayAuthApiKeySnapshot,
@@ -263,6 +269,7 @@ pub(crate) async fn preselect_local_execution_candidates_for_api_formats_with_se
         state,
         client_api_format,
         requested_model,
+        request_operation,
         require_streaming,
         required_capabilities,
         auth_snapshot,
@@ -283,6 +290,7 @@ pub(crate) struct LocalCandidatePreselectionPageCursor<'a> {
     trace_id: String,
     client_api_format: String,
     requested_model: String,
+    request_operation: Option<String>,
     require_streaming: bool,
     required_capabilities: Option<serde_json::Value>,
     auth_snapshot: GatewayAuthApiKeySnapshot,
@@ -336,6 +344,7 @@ impl<'a> LocalCandidatePreselectionPageCursor<'a> {
         model_directive_policy: &crate::system_features::ModelDirectivePolicySnapshot,
         client_api_format: &str,
         requested_model: &str,
+        request_operation: Option<&str>,
         require_streaming: bool,
         required_capabilities: Option<&serde_json::Value>,
         auth_snapshot: &GatewayAuthApiKeySnapshot,
@@ -370,6 +379,7 @@ impl<'a> LocalCandidatePreselectionPageCursor<'a> {
             trace_id: trace_id.unwrap_or_default().to_string(),
             client_api_format: client_api_format.to_string(),
             requested_model: requested_model.to_string(),
+            request_operation: request_operation.map(str::to_string),
             require_streaming,
             required_capabilities: required_capabilities.cloned(),
             auth_snapshot: auth_snapshot.clone(),
@@ -452,6 +462,10 @@ impl<'a> LocalCandidatePreselectionPageCursor<'a> {
         self.key_mode.cache_key_name()
     }
 
+    pub(crate) fn resolved_page_cache_request_operation(&self) -> Option<&str> {
+        self.request_operation.as_deref()
+    }
+
     pub(crate) fn resolved_page_cache_use_api_format_alias_match(&self) -> bool {
         self.use_api_format_alias_match
     }
@@ -515,6 +529,7 @@ impl<'a> LocalCandidatePreselectionPageCursor<'a> {
     > {
         let key = CandidatePageCacheKey::new(
             &self.requested_model,
+            self.request_operation.as_deref(),
             &self.client_api_format,
             self.require_streaming,
             &self.auth_snapshot,
@@ -965,11 +980,12 @@ impl<'a> LocalCandidatePreselectionPageCursor<'a> {
             .map_err(|err| GatewayError::Internal(err.to_string()))?
             .into_iter()
             .filter(|row| {
-                row_supports_requested_model_with_model_directives(
+                row_supports_requested_model_with_model_directives_and_request_operation(
                     row,
                     &routing_model,
                     normalized_api_format,
                     false,
+                    self.request_operation.as_deref(),
                 )
             })
             .collect::<Vec<_>>();
@@ -1009,12 +1025,15 @@ impl<'a> LocalCandidatePreselectionPageCursor<'a> {
             if let Some(value) = self.resolved_global_model_names.get(normalized_api_format) {
                 value.clone()
             } else {
-                let Some(value) = resolve_requested_global_model_name_with_model_directives(
-                    &rows,
-                    &routing_model,
-                    normalized_api_format,
-                    false,
-                ) else {
+                let Some(value) =
+                    resolve_requested_global_model_name_with_model_directives_and_request_operation(
+                        &rows,
+                        &routing_model,
+                        normalized_api_format,
+                        false,
+                        self.request_operation.as_deref(),
+                    )
+                else {
                     return Ok(None);
                 };
                 self.resolved_global_model_names
@@ -1037,6 +1056,7 @@ impl<'a> LocalCandidatePreselectionPageCursor<'a> {
             EnumerateMinimalCandidateSelectionInput {
                 rows,
                 normalized_api_format,
+                request_operation: self.request_operation.as_deref(),
                 requested_model_name: &routing_model,
                 resolved_global_model_name: resolved_global_model_name.as_str(),
                 require_streaming: self.require_streaming,
@@ -1316,6 +1336,7 @@ mod tests {
             &model_directive_policy,
             "openai:chat",
             "gpt-5",
+            None,
             true,
             None,
             &auth_snapshot,
@@ -1531,6 +1552,7 @@ mod tests {
                 priority: 1,
                 api_formats: None,
                 endpoint_ids: Some(vec!["endpoint-opg-openai".to_string()]),
+                operations: None,
             }]),
             model_supports_streaming: Some(true),
             model_is_active: true,
@@ -1561,6 +1583,7 @@ mod tests {
             &model_directive_policy,
             "claude:messages",
             "gpt-5.5-xhigh",
+            None,
             false,
             None,
             &auth_snapshot,
@@ -1590,6 +1613,70 @@ mod tests {
         );
     }
 
+    #[tokio::test]
+    async fn paged_preselection_prefers_operation_scoped_mapping_for_compaction() {
+        let mut row = openai_responses_mapping_row();
+        row.global_model_mappings = None;
+        row.global_model_name = "gpt-5.6-sol".to_string();
+        row.model_provider_model_name = "gpt-5.6-sol".to_string();
+        row.model_provider_model_mappings = Some(vec![
+            StoredProviderModelMapping {
+                name: "gpt-5.6-sol".to_string(),
+                priority: 1,
+                api_formats: Some(vec!["openai:responses".to_string()]),
+                endpoint_ids: None,
+                operations: None,
+            },
+            StoredProviderModelMapping {
+                name: "gpt-5.6-terra".to_string(),
+                priority: 1,
+                api_formats: Some(vec!["openai:responses".to_string()]),
+                endpoint_ids: None,
+                operations: Some(vec!["compact".to_string()]),
+            },
+        ]);
+        let repository: Arc<dyn MinimalCandidateSelectionReadRepository> =
+            Arc::new(InMemoryMinimalCandidateSelectionReadRepository::seed([row]));
+        let data_state =
+            GatewayDataState::with_minimal_candidate_selection_reader_for_tests(repository);
+        let app = AppState::new()
+            .expect("gateway state should build")
+            .with_data_state_for_tests(data_state);
+        let auth_snapshot = unrestricted_auth_snapshot();
+        let model_directive_policy =
+            crate::system_features::ModelDirectivePolicySnapshot::load(&app).await;
+        let mut cursor = LocalCandidatePreselectionPageCursor::new(
+            PlannerAppState::new(&app),
+            &model_directive_policy,
+            "openai:responses",
+            "gpt-5.6-sol",
+            Some("compact"),
+            false,
+            None,
+            &auth_snapshot,
+            None,
+            None,
+            None,
+            true,
+            LocalCandidatePreselectionKeyMode::ProviderEndpointKeyModelAndApiFormat,
+            true,
+            None,
+        )
+        .await;
+
+        let page = cursor
+            .next_page()
+            .await
+            .expect("preselection should succeed")
+            .expect("compact mapping should find a provider");
+
+        assert_eq!(page.candidates.len(), 1);
+        assert_eq!(
+            page.candidates[0].selected_provider_model_name,
+            "gpt-5.6-terra"
+        );
+    }
+
     #[tokio::test]
     async fn custom_policy_suffix_uses_the_same_base_model_for_candidate_selection() {
         let mut row = openai_responses_mapping_row();
@@ -1634,6 +1721,7 @@ mod tests {
             &model_directive_policy,
             "openai:responses",
             "deployment-alias-VendorFuture",
+            None,
             false,
             None,
             &auth_snapshot,
@@ -1695,6 +1783,7 @@ mod tests {
             &model_directive_policy,
             "claude:messages",
             "deepseek-v4-pro",
+            None,
             false,
             None,
             &auth_snapshot,
@@ -1773,6 +1862,7 @@ mod tests {
             &model_directive_policy,
             "claude:messages",
             "gpt-5",
+            None,
             false,
             None,
             &auth_snapshot,
diff --git a/apps/aether-gateway/src/ai_serving/planner/standard/family/candidates.rs b/apps/aether-gateway/src/ai_serving/planner/standard/family/candidates.rs
index 8cea21bd6..ffe92f270 100644
--- a/apps/aether-gateway/src/ai_serving/planner/standard/family/candidates.rs
+++ b/apps/aether-gateway/src/ai_serving/planner/standard/family/candidates.rs
@@ -124,6 +124,7 @@ pub(super) async fn materialize_local_standard_candidate_attempts(
         &input.model_directive_policy,
         spec_metadata.api_format,
         &input.requested_model,
+        None,
         false,
         input.required_capabilities.as_ref(),
         &input.auth_snapshot,
@@ -250,6 +251,7 @@ pub(super) async fn build_local_standard_candidate_attempt_source<'a>(
             trace_id,
             spec_metadata.api_format,
             &input.requested_model,
+            None,
             spec_metadata.require_streaming,
             &input.auth_snapshot,
             input.client_session_affinity.as_ref(),
@@ -345,6 +347,7 @@ async fn maybe_append_gemini_image_openai_image_preselection(
         &input.model_directive_policy,
         spec_metadata.api_format,
         &input.requested_model,
+        None,
         spec_metadata.require_streaming,
         input.required_capabilities.as_ref(),
         &input.auth_snapshot,
diff --git a/apps/aether-gateway/src/ai_serving/planner/standard/openai/chat/decision/support.rs b/apps/aether-gateway/src/ai_serving/planner/standard/openai/chat/decision/support.rs
index f57237ec2..4a61a6484 100644
--- a/apps/aether-gateway/src/ai_serving/planner/standard/openai/chat/decision/support.rs
+++ b/apps/aether-gateway/src/ai_serving/planner/standard/openai/chat/decision/support.rs
@@ -297,6 +297,7 @@ pub(crate) async fn build_lazy_local_openai_chat_candidate_attempt_source<'a>(
         trace_id,
         "openai:chat",
         &input.requested_model,
+        None,
         require_streaming,
         &input.auth_snapshot,
         input.client_session_affinity.as_ref(),
diff --git a/apps/aether-gateway/src/ai_serving/planner/standard/openai/chat/plans/candidates.rs b/apps/aether-gateway/src/ai_serving/planner/standard/openai/chat/plans/candidates.rs
index e0db3b2b1..0b644d19a 100644
--- a/apps/aether-gateway/src/ai_serving/planner/standard/openai/chat/plans/candidates.rs
+++ b/apps/aether-gateway/src/ai_serving/planner/standard/openai/chat/plans/candidates.rs
@@ -24,6 +24,7 @@ pub(crate) async fn list_local_openai_chat_candidates(
         &input.model_directive_policy,
         "openai:chat",
         &input.requested_model,
+        None,
         require_streaming,
         input.required_capabilities.as_ref(),
         &input.auth_snapshot,
diff --git a/apps/aether-gateway/src/ai_serving/planner/standard/openai/responses/decision/support.rs b/apps/aether-gateway/src/ai_serving/planner/standard/openai/responses/decision/support.rs
index c44d3782c..3ad3674c0 100644
--- a/apps/aether-gateway/src/ai_serving/planner/standard/openai/responses/decision/support.rs
+++ b/apps/aether-gateway/src/ai_serving/planner/standard/openai/responses/decision/support.rs
@@ -31,8 +31,8 @@ use crate::ai_serving::planner::spec_metadata::local_openai_responses_spec_metad
 use crate::ai_serving::planner::CandidateFailureDiagnostic;
 use crate::ai_serving::{
     ai_local_execution_contract_for_formats, extract_pool_sticky_session_token,
-    resolve_local_decision_execution_runtime_auth_context, ExecutionRuntimeAuthContext,
-    GatewayControlDecision, PlannerAppState,
+    openai_responses_request_operation, resolve_local_decision_execution_runtime_auth_context,
+    ExecutionRuntimeAuthContext, GatewayControlDecision, PlannerAppState,
 };
 use crate::client_session_affinity::client_session_affinity_from_parts;
 use crate::{AppState, GatewayError};
@@ -163,6 +163,7 @@ pub(crate) async fn materialize_local_openai_responses_candidate_attempts(
     spec: LocalOpenAiResponsesSpec,
 ) -> Result<(Vec<LocalOpenAiResponsesCandidateAttempt>, usize), GatewayError> {
     let spec_metadata = local_openai_responses_spec_metadata(spec);
+    let request_operation = openai_responses_request_operation(spec_metadata.api_format, body_json);
     let planner_state = PlannerAppState::new(state);
     let sticky_session_token = extract_pool_sticky_session_token(body_json);
     let auth_context: &ExecutionRuntimeAuthContext = &input.auth_context;
@@ -176,6 +177,7 @@ pub(crate) async fn materialize_local_openai_responses_candidate_attempts(
         &input.model_directive_policy,
         spec_metadata.api_format,
         &input.requested_model,
+        request_operation,
         spec_metadata.require_streaming,
         input.required_capabilities.as_ref(),
         &input.auth_snapshot,
@@ -262,6 +264,7 @@ pub(crate) async fn build_local_openai_responses_candidate_attempt_source<'a>(
     spec: LocalOpenAiResponsesSpec,
 ) -> Result<(LocalOpenAiResponsesCandidateAttemptSource<'a>, usize), GatewayError> {
     let spec_metadata = local_openai_responses_spec_metadata(spec);
+    let request_operation = openai_responses_request_operation(spec_metadata.api_format, body_json);
     let planner_state = PlannerAppState::new(state);
     let sticky_session_token = extract_pool_sticky_session_token(body_json);
     let auth_context: &ExecutionRuntimeAuthContext = &input.auth_context;
@@ -287,6 +290,7 @@ pub(crate) async fn build_local_openai_responses_candidate_attempt_source<'a>(
             trace_id,
             spec_metadata.api_format,
             &input.requested_model,
+            request_operation,
             spec_metadata.require_streaming,
             &input.auth_snapshot,
             input.client_session_affinity.as_ref(),
@@ -372,6 +376,7 @@ pub(crate) async fn build_local_openai_responses_image_candidate_attempt_source<
         &input.model_directive_policy,
         spec_metadata.api_format,
         &input.requested_model,
+        None,
         false,
         input.required_capabilities.as_ref(),
         &input.auth_snapshot,
diff --git a/apps/aether-gateway/src/ai_serving/planner/state/scheduler.rs b/apps/aether-gateway/src/ai_serving/planner/state/scheduler.rs
index 22571c73b..916307790 100644
--- a/apps/aether-gateway/src/ai_serving/planner/state/scheduler.rs
+++ b/apps/aether-gateway/src/ai_serving/planner/state/scheduler.rs
@@ -53,13 +53,45 @@ impl<'a> PlannerAppState<'a> {
             Vec<SchedulerSkippedCandidate>,
         ),
         GatewayError,
+    > {
+        self.list_selectable_candidates_with_skip_reasons_for_request_operation(
+            api_format,
+            global_model_name,
+            require_streaming,
+            required_capabilities,
+            auth_snapshot,
+            client_session_affinity,
+            now_unix_secs,
+            enable_model_directives,
+            None,
+        )
+        .await
+    }
+
+    pub(crate) async fn list_selectable_candidates_with_skip_reasons_for_request_operation(
+        self,
+        api_format: &str,
+        global_model_name: &str,
+        require_streaming: bool,
+        required_capabilities: Option<&serde_json::Value>,
+        auth_snapshot: Option<&GatewayAuthApiKeySnapshot>,
+        client_session_affinity: Option<&ClientSessionAffinity>,
+        now_unix_secs: u64,
+        enable_model_directives: bool,
+        request_operation: Option<&str>,
+    ) -> Result<
+        (
+            Vec<SchedulerMinimalCandidateSelectionCandidate>,
+            Vec<SchedulerSkippedCandidate>,
+        ),
+        GatewayError,
     > {
         let wait_timeout = Duration::from_millis(API_KEY_CONCURRENCY_WAIT_TIMEOUT_MS);
         let wait_interval = Duration::from_millis(API_KEY_CONCURRENCY_WAIT_POLL_INTERVAL_MS.max(1));
         let wait_deadline = Instant::now() + wait_timeout;
         let mut attempt_now_unix_secs = now_unix_secs;
         loop {
-            let result = crate::scheduler::candidate::list_selectable_candidates_with_skip_reasons(
+            let result = crate::scheduler::candidate::list_selectable_candidates_with_skip_reasons_for_request_operation(
                 self.app().data.as_ref(),
                 self.app(),
                 api_format,
@@ -70,6 +102,7 @@ impl<'a> PlannerAppState<'a> {
                 client_session_affinity,
                 attempt_now_unix_secs,
                 enable_model_directives,
+                request_operation,
             )
             .await?;
 
diff --git a/apps/aether-gateway/src/ai_serving/pure/mod.rs b/apps/aether-gateway/src/ai_serving/pure/mod.rs
index c1a22da71..585ef13ad 100644
--- a/apps/aether-gateway/src/ai_serving/pure/mod.rs
+++ b/apps/aether-gateway/src/ai_serving/pure/mod.rs
@@ -165,5 +165,5 @@ pub(crate) use aether_ai_formats::api::{
 pub(crate) use aether_ai_formats::{
     api_format_defaults_to_client_error_failover, api_format_defaults_to_non_stream,
     api_format_permission_covers, intersect_api_format_allowed_lists, is_embedding_api_format,
-    is_rerank_api_format,
+    is_rerank_api_format, openai_responses_request_operation,
 };
diff --git a/apps/aether-gateway/src/cache/candidate_page.rs b/apps/aether-gateway/src/cache/candidate_page.rs
index 0c22546bf..1b6323e74 100644
--- a/apps/aether-gateway/src/cache/candidate_page.rs
+++ b/apps/aether-gateway/src/cache/candidate_page.rs
@@ -82,6 +82,7 @@ pub(crate) struct CandidateRowPageCacheKey {
 #[derive(Debug, Clone, PartialEq, Eq, Hash)]
 pub(crate) struct CandidatePageCacheKey {
     requested_model: String,
+    request_operation: String,
     client_api_format: String,
     auth_identity: CandidatePageAuthIdentity,
     require_streaming: bool,
@@ -131,6 +132,7 @@ impl CandidatePageCacheKey {
     #[allow(clippy::too_many_arguments)]
     pub(crate) fn new(
         requested_model: &str,
+        request_operation: Option<&str>,
         client_api_format: &str,
         require_streaming: bool,
         auth_snapshot: &GatewayAuthApiKeySnapshot,
@@ -145,6 +147,7 @@ impl CandidatePageCacheKey {
     ) -> Self {
         Self {
             requested_model: normalize_text_key(requested_model),
+            request_operation: normalize_text_key(request_operation.unwrap_or_default()),
             client_api_format: normalize_api_format(client_api_format),
             auth_identity: CandidatePageAuthIdentity::from_auth_snapshot(auth_snapshot),
             require_streaming,
@@ -164,6 +167,7 @@ impl CandidateResolvedPageCacheKey {
     #[allow(clippy::too_many_arguments)]
     pub(crate) fn new(
         requested_model: &str,
+        request_operation: Option<&str>,
         client_api_format: &str,
         require_streaming: bool,
         auth_snapshot: &GatewayAuthApiKeySnapshot,
@@ -180,6 +184,7 @@ impl CandidateResolvedPageCacheKey {
         Self {
             page_key: CandidatePageCacheKey::new(
                 requested_model,
+                request_operation,
                 client_api_format,
                 require_streaming,
                 auth_snapshot,
@@ -564,6 +569,7 @@ mod tests {
         let auth_b = auth_snapshot("user-b", "key-a");
         let base = CandidatePageCacheKey::new(
             "gpt-4o",
+            None,
             "openai:chat",
             true,
             &auth_a,
@@ -578,6 +584,7 @@ mod tests {
         );
         let different_user = CandidatePageCacheKey::new(
             "gpt-4o",
+            None,
             "openai:chat",
             true,
             &auth_b,
@@ -592,6 +599,22 @@ mod tests {
         );
         let different_model = CandidatePageCacheKey::new(
             "gpt-4.1",
+            None,
+            "openai:chat",
+            true,
+            &auth_a,
+            Some(&json!({"vision": true})),
+            None,
+            Some("bearer"),
+            7,
+            "provider_endpoint_key_model",
+            true,
+            None,
+            "policy-a",
+        );
+        let different_operation = CandidatePageCacheKey::new(
+            "gpt-4o",
+            Some("compact"),
             "openai:chat",
             true,
             &auth_a,
@@ -606,6 +629,7 @@ mod tests {
         );
         let different_format = CandidatePageCacheKey::new(
             "gpt-4o",
+            None,
             "openai:responses",
             true,
             &auth_a,
@@ -620,6 +644,7 @@ mod tests {
         );
         let different_capabilities = CandidatePageCacheKey::new(
             "gpt-4o",
+            None,
             "openai:chat",
             true,
             &auth_a,
@@ -634,6 +659,7 @@ mod tests {
         );
         let same_policy = CandidatePageCacheKey::new(
             "gpt-4o",
+            None,
             "openai:chat",
             true,
             &auth_a,
@@ -648,6 +674,7 @@ mod tests {
         );
         let different_policy = CandidatePageCacheKey::new(
             "gpt-4o",
+            None,
             "openai:chat",
             true,
             &auth_a,
@@ -664,12 +691,14 @@ mod tests {
         assert_eq!(base, same_policy);
         assert_ne!(base, different_user);
         assert_ne!(base, different_model);
+        assert_ne!(base, different_operation);
         assert_ne!(base, different_format);
         assert_ne!(base, different_capabilities);
         assert_ne!(base, different_policy);
 
         let resolved_base = CandidateResolvedPageCacheKey::new(
             "gpt-4o",
+            None,
             "openai:chat",
             true,
             &auth_a,
@@ -685,6 +714,7 @@ mod tests {
         );
         let resolved_same_policy = CandidateResolvedPageCacheKey::new(
             "gpt-4o",
+            None,
             "openai:chat",
             true,
             &auth_a,
@@ -700,6 +730,7 @@ mod tests {
         );
         let resolved_different_policy = CandidateResolvedPageCacheKey::new(
             "gpt-4o",
+            None,
             "openai:chat",
             true,
             &auth_a,
diff --git a/apps/aether-gateway/src/control/auth/gate.rs b/apps/aether-gateway/src/control/auth/gate.rs
index 3a63bad65..047432688 100644
--- a/apps/aether-gateway/src/control/auth/gate.rs
+++ b/apps/aether-gateway/src/control/auth/gate.rs
@@ -610,6 +610,7 @@ mod tests {
                 priority: 1,
                 api_formats: Some(vec!["openai:chat".to_string()]),
                 endpoint_ids: None,
+                operations: None,
             }]),
             model_supports_streaming: Some(true),
             model_is_active: true,
@@ -905,6 +906,7 @@ mod tests {
             priority: 1,
             api_formats: Some(vec!["gemini:generate_content".to_string()]),
             endpoint_ids: None,
+            operations: None,
         }]);
         let state = state_with_rows(vec![row]);
         let decision = decision_with_allowed_models(vec!["gpt-5".to_string()]);
diff --git a/apps/aether-gateway/src/data/candidate_selection.rs b/apps/aether-gateway/src/data/candidate_selection.rs
index a8a5d794b..003a61661 100644
--- a/apps/aether-gateway/src/data/candidate_selection.rs
+++ b/apps/aether-gateway/src/data/candidate_selection.rs
@@ -239,6 +239,29 @@ pub(crate) async fn enumerate_minimal_candidate_selection_with_required_capabili
     auth_snapshot: Option<&GatewayAuthApiKeySnapshot>,
     required_capabilities: Option<&serde_json::Value>,
     enable_model_directives: bool,
+) -> Result<Vec<SchedulerMinimalCandidateSelectionCandidate>, DataLayerError> {
+    enumerate_minimal_candidate_selection_with_required_capabilities_for_request_operation(
+        state,
+        api_format,
+        requested_model_name,
+        require_streaming,
+        auth_snapshot,
+        required_capabilities,
+        enable_model_directives,
+        None,
+    )
+    .await
+}
+
+pub(crate) async fn enumerate_minimal_candidate_selection_with_required_capabilities_for_request_operation(
+    state: &(impl MinimalCandidateSelectionRowSource + Sync),
+    api_format: &str,
+    requested_model_name: &str,
+    require_streaming: bool,
+    auth_snapshot: Option<&GatewayAuthApiKeySnapshot>,
+    required_capabilities: Option<&serde_json::Value>,
+    enable_model_directives: bool,
+    request_operation: Option<&str>,
 ) -> Result<Vec<SchedulerMinimalCandidateSelectionCandidate>, DataLayerError> {
     let normalized_api_format = normalize_api_format(api_format);
     if normalized_api_format.is_empty() {
@@ -267,6 +290,7 @@ pub(crate) async fn enumerate_minimal_candidate_selection_with_required_capabili
         EnumerateMinimalCandidateSelectionInput {
             rows,
             normalized_api_format: &normalized_api_format,
+            request_operation,
             requested_model_name,
             resolved_global_model_name: resolved_global_model_name.as_str(),
             require_streaming,
diff --git a/apps/aether-gateway/src/data/tests.rs b/apps/aether-gateway/src/data/tests.rs
index 12078c90e..f851fa79b 100644
--- a/apps/aether-gateway/src/data/tests.rs
+++ b/apps/aether-gateway/src/data/tests.rs
@@ -588,6 +588,7 @@ fn sample_minimal_candidate_selection_row(
             priority: 1,
             api_formats: Some(vec!["openai:chat".to_string()]),
             endpoint_ids: None,
+            operations: None,
         }]),
         model_supports_streaming: None,
         model_is_active: true,
@@ -905,6 +906,7 @@ async fn data_state_reads_minimal_candidate_selection_with_auth_filters() {
         enumerate_minimal_candidate_selection(EnumerateMinimalCandidateSelectionInput {
             rows,
             normalized_api_format: "openai:chat",
+            request_operation: None,
             requested_model_name: "gpt-4.1",
             resolved_global_model_name: "gpt-4.1",
             require_streaming: false,
diff --git a/apps/aether-gateway/src/handlers/admin/provider/query/models/model_test/model_mapping.rs b/apps/aether-gateway/src/handlers/admin/provider/query/models/model_test/model_mapping.rs
index e4f670e40..ed456959d 100644
--- a/apps/aether-gateway/src/handlers/admin/provider/query/models/model_test/model_mapping.rs
+++ b/apps/aether-gateway/src/handlers/admin/provider/query/models/model_test/model_mapping.rs
@@ -213,6 +213,7 @@ fn provider_query_parse_embedded_provider_model_mappings(
         priority: 1,
         api_formats: None,
         endpoint_ids: None,
+        operations: None,
     }]))
 }
 
@@ -237,6 +238,7 @@ fn provider_query_parse_provider_model_mappings_array(
                         priority: 1,
                         api_formats: None,
                         endpoint_ids: None,
+                        operations: None,
                     });
                 }
             }
@@ -293,6 +295,11 @@ fn provider_query_parse_provider_model_mapping_object_lenient(
         object.get("endpoint_ids"),
         "models.provider_model_mappings.endpoint_ids",
     )?;
+    let operations = provider_query_parse_mapping_string_list(
+        object.get("operations"),
+        "models.provider_model_mappings.operations",
+    )?
+    .and_then(provider_query_normalize_request_operations);
 
     Ok(Some(StoredProviderModelMapping {
         name: name.to_string(),
@@ -303,9 +310,19 @@ fn provider_query_parse_provider_model_mapping_object_lenient(
         })?,
         api_formats,
         endpoint_ids,
+        operations,
     }))
 }
 
+fn provider_query_normalize_request_operations(values: Vec<String>) -> Option<Vec<String>> {
+    let operations = values
+        .into_iter()
+        .map(|value| value.trim().to_ascii_lowercase())
+        .filter(|value| !value.is_empty())
+        .collect::<Vec<_>>();
+    (!operations.is_empty()).then_some(operations)
+}
+
 fn provider_query_parse_mapping_string_list(
     value: Option<&Value>,
     field_name: &str,
@@ -389,6 +406,7 @@ mod tests {
             priority: 1,
             api_formats: Some(vec![api_format.to_string()]),
             endpoint_ids: None,
+            operations: None,
         }
     }
 
diff --git a/apps/aether-gateway/src/handlers/admin/request/models.rs b/apps/aether-gateway/src/handlers/admin/request/models.rs
index 1da58be85..b24daf716 100644
--- a/apps/aether-gateway/src/handlers/admin/request/models.rs
+++ b/apps/aether-gateway/src/handlers/admin/request/models.rs
@@ -15,7 +15,7 @@ use serde_json::json;
 use std::collections::{BTreeMap, BTreeSet};
 use uuid::Uuid;
 
-fn normalize_provider_model_mappings_api_formats(
+fn normalize_provider_model_mapping_scopes(
     value: Option<serde_json::Value>,
 ) -> Option<serde_json::Value> {
     let Some(mut value) = value else {
@@ -36,6 +36,9 @@ fn normalize_provider_model_mappings_api_formats(
         normalize_provider_model_mapping_string_array_field(object, "endpoint_ids", |value| {
             value.trim().to_string()
         });
+        normalize_provider_model_mapping_string_array_field(object, "operations", |value| {
+            value.trim().to_ascii_lowercase()
+        });
     }
     Some(value)
 }
@@ -155,7 +158,7 @@ impl<'a> AdminAppState<'a> {
             "price_per_request",
         )?;
         let tiered_pricing = normalize_json_object(payload.tiered_pricing, "tiered_pricing")?;
-        let provider_model_mappings = normalize_provider_model_mappings_api_formats(
+        let provider_model_mappings = normalize_provider_model_mapping_scopes(
             normalize_json_array(payload.provider_model_mappings, "provider_model_mappings")?,
         );
         let config = normalize_json_object(payload.config, "config")?;
@@ -241,7 +244,7 @@ impl<'a> AdminAppState<'a> {
             existing.tiered_pricing.clone()
         };
         let provider_model_mappings = if fields.contains("provider_model_mappings") {
-            normalize_provider_model_mappings_api_formats(normalize_json_array(
+            normalize_provider_model_mapping_scopes(normalize_json_array(
                 payload.provider_model_mappings,
                 "provider_model_mappings",
             )?)
diff --git a/apps/aether-gateway/src/handlers/public/support/user_me_usage.rs b/apps/aether-gateway/src/handlers/public/support/user_me_usage.rs
index 5c63cef02..e66900885 100644
--- a/apps/aether-gateway/src/handlers/public/support/user_me_usage.rs
+++ b/apps/aether-gateway/src/handlers/public/support/user_me_usage.rs
@@ -488,6 +488,7 @@ fn build_users_me_usage_record_payload(
         "cache_read_input_tokens": item.cache_read_input_tokens,
         "status_code": item.status_code,
         "error_message": item.error_message,
+        "request_type": item.request_type,
         "input_price_per_1m": input_price_per_1m,
         "output_price_per_1m": output_price_per_1m,
         "cache_creation_price_per_1m": cache_creation_price_per_1m,
@@ -525,6 +526,7 @@ fn build_users_me_usage_active_payload(item: &StoredRequestUsageAudit) -> serde_
     let mut payload = json!({
         "id": item.id,
         "status": item.status,
+        "request_type": item.request_type,
         "input_tokens": item.input_tokens,
         "effective_input_tokens": users_me_usage_effective_input_tokens(item),
         "output_tokens": item.output_tokens,
diff --git a/apps/aether-gateway/src/scheduler/candidate/enumeration.rs b/apps/aether-gateway/src/scheduler/candidate/enumeration.rs
index c04fc1cfe..12a6f4f21 100644
--- a/apps/aether-gateway/src/scheduler/candidate/enumeration.rs
+++ b/apps/aether-gateway/src/scheduler/candidate/enumeration.rs
@@ -2,7 +2,7 @@ use aether_scheduler_core::SchedulerMinimalCandidateSelectionCandidate;
 
 use crate::data::auth::GatewayAuthApiKeySnapshot;
 use crate::data::candidate_selection::{
-    enumerate_minimal_candidate_selection_with_required_capabilities,
+    enumerate_minimal_candidate_selection_with_required_capabilities_for_request_operation,
     MinimalCandidateSelectionRowSource,
 };
 use crate::GatewayError;
@@ -15,8 +15,9 @@ pub(super) async fn enumerate_scheduler_candidates(
     required_capabilities: Option<&serde_json::Value>,
     auth_snapshot: Option<&GatewayAuthApiKeySnapshot>,
     enable_model_directives: bool,
+    request_operation: Option<&str>,
 ) -> Result<Vec<SchedulerMinimalCandidateSelectionCandidate>, GatewayError> {
-    enumerate_minimal_candidate_selection_with_required_capabilities(
+    enumerate_minimal_candidate_selection_with_required_capabilities_for_request_operation(
         selection_row_source,
         api_format,
         global_model_name,
@@ -24,6 +25,7 @@ pub(super) async fn enumerate_scheduler_candidates(
         auth_snapshot,
         required_capabilities,
         enable_model_directives,
+        request_operation,
     )
     .await
     .map_err(|err| GatewayError::Internal(err.to_string()))
diff --git a/apps/aether-gateway/src/scheduler/candidate/mod.rs b/apps/aether-gateway/src/scheduler/candidate/mod.rs
index f793cef21..1da74c2ad 100644
--- a/apps/aether-gateway/src/scheduler/candidate/mod.rs
+++ b/apps/aether-gateway/src/scheduler/candidate/mod.rs
@@ -116,6 +116,42 @@ pub(crate) async fn list_selectable_candidates_with_skip_reasons(
         client_session_affinity,
         now_unix_secs,
         enable_model_directives,
+        None,
+    )
+    .await
+}
+
+pub(crate) async fn list_selectable_candidates_with_skip_reasons_for_request_operation(
+    selection_row_source: &(impl MinimalCandidateSelectionRowSource + Sync),
+    runtime_state: &impl SchedulerRuntimeState,
+    api_format: &str,
+    global_model_name: &str,
+    require_streaming: bool,
+    required_capabilities: Option<&serde_json::Value>,
+    auth_snapshot: Option<&GatewayAuthApiKeySnapshot>,
+    client_session_affinity: Option<&ClientSessionAffinity>,
+    now_unix_secs: u64,
+    enable_model_directives: bool,
+    request_operation: Option<&str>,
+) -> Result<
+    (
+        Vec<SchedulerMinimalCandidateSelectionCandidate>,
+        Vec<SchedulerSkippedCandidate>,
+    ),
+    GatewayError,
+> {
+    collect_selectable_candidates_with_skip_reasons(
+        selection_row_source,
+        runtime_state,
+        api_format,
+        global_model_name,
+        require_streaming,
+        required_capabilities,
+        auth_snapshot,
+        client_session_affinity,
+        now_unix_secs,
+        enable_model_directives,
+        request_operation,
     )
     .await
 }
@@ -237,6 +273,7 @@ pub(crate) async fn list_selectable_candidates_for_required_capability_without_r
             client_session_affinity,
             now_unix_secs,
             false,
+            None,
         )
         .await?;
         all_attempts_blocked_by_auth_limit &=
diff --git a/apps/aether-gateway/src/scheduler/candidate/selection.rs b/apps/aether-gateway/src/scheduler/candidate/selection.rs
index 7ef674b11..5a784dced 100644
--- a/apps/aether-gateway/src/scheduler/candidate/selection.rs
+++ b/apps/aether-gateway/src/scheduler/candidate/selection.rs
@@ -81,6 +81,7 @@ pub(super) async fn select_minimal_candidate(
         required_capabilities,
         auth_snapshot,
         enable_model_directives,
+        None,
     )
     .await?;
     let selected = collect_selectable_enumerated_candidates_with_skip_reasons(
@@ -137,6 +138,7 @@ pub(super) async fn collect_selectable_candidates(
         client_session_affinity,
         now_unix_secs,
         enable_model_directives,
+        None,
     )
     .await?
     .0)
@@ -153,6 +155,7 @@ pub(super) async fn collect_selectable_candidates_with_skip_reasons(
     client_session_affinity: Option<&ClientSessionAffinity>,
     now_unix_secs: u64,
     enable_model_directives: bool,
+    request_operation: Option<&str>,
 ) -> Result<
     (
         Vec<SchedulerMinimalCandidateSelectionCandidate>,
@@ -174,6 +177,7 @@ pub(super) async fn collect_selectable_candidates_with_skip_reasons(
         required_capabilities,
         auth_snapshot,
         enable_model_directives,
+        request_operation,
     )
     .await?;
     collect_selectable_enumerated_candidates_with_skip_reasons(
diff --git a/apps/aether-gateway/src/scheduler/candidate/tests/affinity.rs b/apps/aether-gateway/src/scheduler/candidate/tests/affinity.rs
index 50facf3dc..0fe2b2467 100644
--- a/apps/aether-gateway/src/scheduler/candidate/tests/affinity.rs
+++ b/apps/aether-gateway/src/scheduler/candidate/tests/affinity.rs
@@ -107,6 +107,7 @@ async fn same_priority_candidates_are_distributed_by_affinity_key() {
         priority: 1,
         api_formats: Some(vec!["openai:chat".to_string()]),
         endpoint_ids: None,
+        operations: None,
     }]);
 
     let mut second = sample_row();
@@ -124,6 +125,7 @@ async fn same_priority_candidates_are_distributed_by_affinity_key() {
         priority: 1,
         api_formats: Some(vec!["openai:chat".to_string()]),
         endpoint_ids: None,
+        operations: None,
     }]);
 
     let candidates = Arc::new(InMemoryMinimalCandidateSelectionReadRepository::seed(vec![
@@ -142,6 +144,7 @@ async fn same_priority_candidates_are_distributed_by_affinity_key() {
         enumerate_minimal_candidate_selection(EnumerateMinimalCandidateSelectionInput {
             rows,
             normalized_api_format: "openai:chat",
+            request_operation: None,
             requested_model_name: "gpt-4.1",
             resolved_global_model_name: "gpt-4.1",
             require_streaming: false,
diff --git a/apps/aether-gateway/src/scheduler/candidate/tests/model.rs b/apps/aether-gateway/src/scheduler/candidate/tests/model.rs
index 94b79ed2c..cfe925f84 100644
--- a/apps/aether-gateway/src/scheduler/candidate/tests/model.rs
+++ b/apps/aether-gateway/src/scheduler/candidate/tests/model.rs
@@ -45,6 +45,7 @@ fn provider_model_mapping_respects_endpoint_scope() {
         priority: 1,
         api_formats: Some(vec!["openai:chat".to_string()]),
         endpoint_ids: Some(vec!["endpoint-selected".to_string()]),
+        operations: None,
     }]);
 
     assert_eq!(
@@ -101,6 +102,7 @@ fn resolves_requested_global_model_from_provider_model_alias() {
         priority: 1,
         api_formats: Some(vec!["openai:chat".to_string()]),
         endpoint_ids: None,
+        operations: None,
     }]);
 
     let resolved = resolve_requested_global_model_name(&[row], "gpt-5.2", "openai:chat");
@@ -156,6 +158,7 @@ async fn enumerate_minimal_candidate_selection_resolves_provider_model_alias() {
         priority: 1,
         api_formats: Some(vec!["openai:chat".to_string()]),
         endpoint_ids: None,
+        operations: None,
     }]);
 
     let candidates = Arc::new(InMemoryMinimalCandidateSelectionReadRepository::seed(vec![
@@ -193,6 +196,7 @@ async fn enumerate_minimal_candidate_selection_filters_endpoint_scoped_alias_row
         priority: 1,
         api_formats: Some(vec!["openai:chat".to_string()]),
         endpoint_ids: Some(vec!["endpoint-selected".to_string()]),
+        operations: None,
     }]);
 
     let mut other = selected.clone();
@@ -248,6 +252,7 @@ async fn enumerate_minimal_candidate_selection_keeps_only_resolved_global_model_
         priority: 1,
         api_formats: Some(vec!["openai:chat".to_string()]),
         endpoint_ids: None,
+        operations: None,
     }]);
 
     let candidates = Arc::new(InMemoryMinimalCandidateSelectionReadRepository::seed(vec![
@@ -287,6 +292,7 @@ async fn enumerate_minimal_candidate_selection_allows_resolved_global_model_in_a
         priority: 1,
         api_formats: Some(vec!["openai:chat".to_string()]),
         endpoint_ids: None,
+        operations: None,
     }]);
 
     let candidates = Arc::new(InMemoryMinimalCandidateSelectionReadRepository::seed(vec![
diff --git a/apps/aether-gateway/src/scheduler/candidate/tests/selection.rs b/apps/aether-gateway/src/scheduler/candidate/tests/selection.rs
index 824964199..b1578c777 100644
--- a/apps/aether-gateway/src/scheduler/candidate/tests/selection.rs
+++ b/apps/aether-gateway/src/scheduler/candidate/tests/selection.rs
@@ -105,6 +105,7 @@ async fn collect_selectable_candidates_with_skip_reasons(
         None,
         now_unix_secs,
         false,
+        None,
     )
     .await
 }
@@ -843,6 +844,7 @@ async fn selects_next_candidate_when_first_provider_quota_is_exhausted() {
         priority: 1,
         api_formats: Some(vec!["openai:chat".to_string()]),
         endpoint_ids: None,
+        operations: None,
     }]);
     first.key_global_priority_by_format = Some(serde_json::json!({"openai:chat": 1}));
 
@@ -858,6 +860,7 @@ async fn selects_next_candidate_when_first_provider_quota_is_exhausted() {
         priority: 1,
         api_formats: Some(vec!["openai:chat".to_string()]),
         endpoint_ids: None,
+        operations: None,
     }]);
     second.key_global_priority_by_format = Some(serde_json::json!({"openai:chat": 2}));
 
@@ -914,6 +917,7 @@ async fn cooled_down_when_recent_failures_are_recorded_for_same_key() {
         priority: 1,
         api_formats: Some(vec!["openai:chat".to_string()]),
         endpoint_ids: None,
+        operations: None,
     }]);
     first.key_global_priority_by_format = Some(serde_json::json!({"openai:chat": 1}));
 
@@ -929,6 +933,7 @@ async fn cooled_down_when_recent_failures_are_recorded_for_same_key() {
         priority: 1,
         api_formats: Some(vec!["openai:chat".to_string()]),
         endpoint_ids: None,
+        operations: None,
     }]);
     second.key_global_priority_by_format = Some(serde_json::json!({"openai:chat": 2}));
 
diff --git a/apps/aether-gateway/src/scheduler/candidate/tests/support.rs b/apps/aether-gateway/src/scheduler/candidate/tests/support.rs
index e1e17dd78..1bf9e6f13 100644
--- a/apps/aether-gateway/src/scheduler/candidate/tests/support.rs
+++ b/apps/aether-gateway/src/scheduler/candidate/tests/support.rs
@@ -40,12 +40,14 @@ pub(super) fn sample_row() -> StoredMinimalCandidateSelectionRow {
                 priority: 1,
                 api_formats: Some(vec!["openai:chat".to_string()]),
                 endpoint_ids: None,
+                operations: None,
             },
             StoredProviderModelMapping {
                 name: "gpt-4.1-responses".to_string(),
                 priority: 1,
                 api_formats: Some(vec!["openai:responses".to_string()]),
                 endpoint_ids: None,
+                operations: None,
             },
         ]),
         model_supports_streaming: None,
diff --git a/apps/aether-gateway/src/state/catalog.rs b/apps/aether-gateway/src/state/catalog.rs
index 1e3d5a823..b61fa3f4c 100644
--- a/apps/aether-gateway/src/state/catalog.rs
+++ b/apps/aether-gateway/src/state/catalog.rs
@@ -1404,6 +1404,7 @@ mod tests {
         let ttl = Duration::from_secs(300);
         let cache_key = CandidatePageCacheKey::new(
             "gpt-5",
+            None,
             "openai:chat",
             true,
             &sample_auth_snapshot(),
diff --git a/apps/aether-gateway/src/testkit.rs b/apps/aether-gateway/src/testkit.rs
index 98f74f5a5..04d40c6e8 100644
--- a/apps/aether-gateway/src/testkit.rs
+++ b/apps/aether-gateway/src/testkit.rs
@@ -247,6 +247,7 @@ fn openai_chat_pressure_candidates(
                 priority: 1,
                 api_formats: Some(vec!["openai:chat".to_string()]),
                 endpoint_ids: Some(vec![pressure_endpoint_id(index)]),
+                operations: None,
             }]),
             model_supports_streaming: Some(true),
             model_is_active: true,
diff --git a/apps/aether-gateway/src/tests/ai_execute/finalize_local.rs b/apps/aether-gateway/src/tests/ai_execute/finalize_local.rs
index 7a787497c..c2103c842 100644
--- a/apps/aether-gateway/src/tests/ai_execute/finalize_local.rs
+++ b/apps/aether-gateway/src/tests/ai_execute/finalize_local.rs
@@ -138,6 +138,7 @@ async fn gateway_executes_openai_chat_sync_upstream_stream_via_local_finalize_re
                 priority: 1,
                 api_formats: Some(vec!["openai:chat".to_string()]),
                 endpoint_ids: None,
+                operations: None,
             }]),
             model_supports_streaming: Some(true),
             model_is_active: true,
@@ -637,6 +638,7 @@ async fn gateway_executes_openai_chat_cross_format_upstream_stream_via_local_fin
                 priority: 1,
                 api_formats: Some(vec!["gemini:generate_content".to_string()]),
                 endpoint_ids: None,
+                operations: None,
             }]),
             model_supports_streaming: Some(true),
             model_is_active: true,
@@ -1082,6 +1084,7 @@ async fn gateway_executes_openai_chat_cross_format_tool_use_upstream_stream_via_
                 priority: 1,
                 api_formats: Some(vec!["claude:messages".to_string()]),
                 endpoint_ids: None,
+                operations: None,
             }]),
             model_supports_streaming: Some(true),
             model_is_active: true,
@@ -1564,6 +1567,7 @@ async fn gateway_executes_openai_chat_antigravity_cross_format_sync_via_local_fi
                 priority: 1,
                 api_formats: Some(vec!["gemini:generate_content".to_string()]),
                 endpoint_ids: None,
+                operations: None,
             }]),
             model_supports_streaming: Some(true),
             model_is_active: true,
@@ -2074,6 +2078,7 @@ async fn gateway_executes_openai_chat_cross_format_claude_upstream_sync_via_loca
                 priority: 1,
                 api_formats: Some(vec!["claude:messages".to_string()]),
                 endpoint_ids: None,
+                operations: None,
             }]),
             model_supports_streaming: Some(true),
             model_is_active: true,
@@ -2433,6 +2438,7 @@ async fn gateway_executes_openai_chat_cross_format_gemini_upstream_sync_via_loca
                 priority: 1,
                 api_formats: Some(vec!["gemini:generate_content".to_string()]),
                 endpoint_ids: None,
+                operations: None,
             }]),
             model_supports_streaming: Some(true),
             model_is_active: true,
diff --git a/apps/aether-gateway/src/tests/ai_execute/finalize_local_cli/compact.rs b/apps/aether-gateway/src/tests/ai_execute/finalize_local_cli/compact.rs
index 58f5fe141..2232af5ca 100644
--- a/apps/aether-gateway/src/tests/ai_execute/finalize_local_cli/compact.rs
+++ b/apps/aether-gateway/src/tests/ai_execute/finalize_local_cli/compact.rs
@@ -113,6 +113,7 @@ async fn gateway_executes_openai_responses_compact_openai_family_upstream_stream
                 priority: 1,
                 api_formats: Some(vec!["openai:responses:compact".to_string()]),
                 endpoint_ids: None,
+                operations: None,
             }]),
             model_supports_streaming: Some(true),
             model_is_active: true,
diff --git a/apps/aether-gateway/src/tests/ai_execute/finalize_local_cli/cross_format.rs b/apps/aether-gateway/src/tests/ai_execute/finalize_local_cli/cross_format.rs
index af570cfbd..03471fe8b 100644
--- a/apps/aether-gateway/src/tests/ai_execute/finalize_local_cli/cross_format.rs
+++ b/apps/aether-gateway/src/tests/ai_execute/finalize_local_cli/cross_format.rs
@@ -113,6 +113,7 @@ async fn gateway_executes_openai_responses_cross_format_upstream_stream_via_loca
                 priority: 1,
                 api_formats: Some(vec!["gemini:generate_content".to_string()]),
                 endpoint_ids: None,
+                operations: None,
             }]),
             model_supports_streaming: Some(true),
             model_is_active: true,
@@ -586,6 +587,7 @@ async fn gateway_executes_openai_responses_cross_format_function_call_upstream_s
                 priority: 1,
                 api_formats: Some(vec!["gemini:generate_content".to_string()]),
                 endpoint_ids: None,
+                operations: None,
             }]),
             model_supports_streaming: Some(true),
             model_is_active: true,
@@ -1071,6 +1073,7 @@ async fn gateway_executes_openai_responses_antigravity_cross_format_upstream_str
                 priority: 1,
                 api_formats: Some(vec!["gemini:generate_content".to_string()]),
                 endpoint_ids: None,
+                operations: None,
             }]),
             model_supports_streaming: Some(true),
             model_is_active: true,
diff --git a/apps/aether-gateway/src/tests/ai_execute/finalize_local_cli/direct.rs b/apps/aether-gateway/src/tests/ai_execute/finalize_local_cli/direct.rs
index 7a35e3dde..429e52774 100644
--- a/apps/aether-gateway/src/tests/ai_execute/finalize_local_cli/direct.rs
+++ b/apps/aether-gateway/src/tests/ai_execute/finalize_local_cli/direct.rs
@@ -135,6 +135,7 @@ async fn gateway_executes_openai_responses_sync_upstream_stream_via_local_finali
                 priority: 1,
                 api_formats: Some(vec!["openai:responses".to_string()]),
                 endpoint_ids: None,
+                operations: None,
             }]),
             model_supports_streaming: Some(true),
             model_is_active: true,
@@ -650,6 +651,7 @@ async fn gateway_executes_kiro_claude_cli_sync_upstream_stream_via_local_finaliz
                 priority: 1,
                 api_formats: Some(vec!["claude:messages".to_string()]),
                 endpoint_ids: None,
+                operations: None,
             }]),
             model_supports_streaming: Some(true),
             model_is_active: true,
diff --git a/apps/aether-gateway/src/tests/ai_execute/finalize_local_provider/claude.rs b/apps/aether-gateway/src/tests/ai_execute/finalize_local_provider/claude.rs
index 101a2b829..5991495b0 100644
--- a/apps/aether-gateway/src/tests/ai_execute/finalize_local_provider/claude.rs
+++ b/apps/aether-gateway/src/tests/ai_execute/finalize_local_provider/claude.rs
@@ -131,6 +131,7 @@ async fn gateway_executes_claude_chat_sync_same_format_via_local_finalize_respon
                 priority: 1,
                 api_formats: Some(vec!["claude:messages".to_string()]),
                 endpoint_ids: None,
+                operations: None,
             }]),
             model_supports_streaming: Some(true),
             model_is_active: true,
@@ -587,6 +588,7 @@ async fn gateway_executes_claude_chat_sync_upstream_stream_via_local_finalize_re
                 priority: 1,
                 api_formats: Some(vec!["claude:messages".to_string()]),
                 endpoint_ids: None,
+                operations: None,
             }]),
             model_supports_streaming: Some(true),
             model_is_active: true,
@@ -1048,6 +1050,7 @@ async fn gateway_executes_claude_cli_sync_upstream_stream_via_local_finalize_res
                 priority: 1,
                 api_formats: Some(vec!["claude:messages".to_string()]),
                 endpoint_ids: None,
+                operations: None,
             }]),
             model_supports_streaming: Some(true),
             model_is_active: true,
diff --git a/apps/aether-gateway/src/tests/ai_execute/finalize_local_provider/gemini.rs b/apps/aether-gateway/src/tests/ai_execute/finalize_local_provider/gemini.rs
index 0a2c2ed67..8fa7f5745 100644
--- a/apps/aether-gateway/src/tests/ai_execute/finalize_local_provider/gemini.rs
+++ b/apps/aether-gateway/src/tests/ai_execute/finalize_local_provider/gemini.rs
@@ -136,6 +136,7 @@ async fn gateway_executes_gemini_chat_sync_same_format_via_local_finalize_respon
                 priority: 1,
                 api_formats: Some(vec!["gemini:generate_content".to_string()]),
                 endpoint_ids: None,
+                operations: None,
             }]),
             model_supports_streaming: Some(true),
             model_is_active: true,
@@ -640,6 +641,7 @@ async fn gateway_executes_gemini_chat_sync_upstream_stream_via_local_finalize_re
                 priority: 1,
                 api_formats: Some(vec!["gemini:generate_content".to_string()]),
                 endpoint_ids: None,
+                operations: None,
             }]),
             model_supports_streaming: Some(true),
             model_is_active: true,
@@ -1134,6 +1136,7 @@ async fn gateway_executes_gemini_cli_sync_upstream_stream_via_local_finalize_res
                 priority: 1,
                 api_formats: Some(vec!["gemini:generate_content".to_string()]),
                 endpoint_ids: None,
+                operations: None,
             }]),
             model_supports_streaming: Some(true),
             model_is_active: true,
@@ -1642,6 +1645,7 @@ async fn gateway_executes_antigravity_gemini_cli_sync_upstream_stream_via_local_
                 priority: 1,
                 api_formats: Some(vec!["gemini:generate_content".to_string()]),
                 endpoint_ids: None,
+                operations: None,
             }]),
             model_supports_streaming: Some(true),
             model_is_active: true,
diff --git a/apps/aether-gateway/src/tests/ai_execute/lifecycle.rs b/apps/aether-gateway/src/tests/ai_execute/lifecycle.rs
index e91578634..c3cae80b1 100644
--- a/apps/aether-gateway/src/tests/ai_execute/lifecycle.rs
+++ b/apps/aether-gateway/src/tests/ai_execute/lifecycle.rs
@@ -111,6 +111,7 @@ fn sample_local_openai_candidate_row() -> StoredMinimalCandidateSelectionRow {
             priority: 1,
             api_formats: Some(vec!["openai:chat".to_string()]),
             endpoint_ids: None,
+            operations: None,
         }]),
         model_supports_streaming: Some(true),
         model_is_active: true,
diff --git a/apps/aether-gateway/src/tests/ai_execute/stream/decision.rs b/apps/aether-gateway/src/tests/ai_execute/stream/decision.rs
index cdb8346cb..3f901976d 100644
--- a/apps/aether-gateway/src/tests/ai_execute/stream/decision.rs
+++ b/apps/aether-gateway/src/tests/ai_execute/stream/decision.rs
@@ -131,6 +131,7 @@ async fn gateway_executes_openai_chat_stream_via_local_decision_gate_without_exe
                 priority: 1,
                 api_formats: Some(vec!["openai:chat".to_string()]),
                 endpoint_ids: None,
+                operations: None,
             }]),
             model_supports_streaming: Some(true),
             model_is_active: true,
@@ -535,6 +536,7 @@ async fn gateway_executes_openai_chat_stream_via_local_openai_responses_cross_fo
                 priority: 1,
                 api_formats: Some(vec!["openai:responses".to_string()]),
                 endpoint_ids: None,
+                operations: None,
             }]),
             model_supports_streaming: Some(true),
             model_is_active: true,
@@ -1020,6 +1022,7 @@ async fn gateway_executes_openai_chat_stream_via_local_cross_format_gemini_candi
                 priority: 1,
                 api_formats: Some(vec!["gemini:generate_content".to_string()]),
                 endpoint_ids: None,
+                operations: None,
             }]),
             model_supports_streaming: Some(true),
             model_is_active: true,
@@ -1458,6 +1461,7 @@ async fn gateway_executes_openai_chat_stream_with_custom_path_via_local_decision
                 priority: 1,
                 api_formats: Some(vec!["openai:chat".to_string()]),
                 endpoint_ids: None,
+                operations: None,
             }]),
             model_supports_streaming: Some(true),
             model_is_active: true,
@@ -1961,6 +1965,7 @@ async fn gateway_retries_next_local_openai_chat_stream_candidate_after_retryable
                 priority: 1,
                 api_formats: Some(vec!["openai:chat".to_string()]),
                 endpoint_ids: None,
+                operations: None,
             }]),
             model_supports_streaming: Some(true),
             model_is_active: true,
diff --git a/apps/aether-gateway/src/tests/ai_execute/stream/image.rs b/apps/aether-gateway/src/tests/ai_execute/stream/image.rs
index 3ab8b0422..8ede08204 100644
--- a/apps/aether-gateway/src/tests/ai_execute/stream/image.rs
+++ b/apps/aether-gateway/src/tests/ai_execute/stream/image.rs
@@ -131,6 +131,7 @@ async fn gateway_executes_codex_image_stream_via_local_decision_gate_after_oauth
                 priority: 1,
                 api_formats: Some(vec!["openai:image".to_string()]),
                 endpoint_ids: None,
+                operations: None,
             }]),
             model_supports_streaming: Some(true),
             model_is_active: true,
@@ -517,6 +518,7 @@ async fn gateway_bridges_codex_image_sync_json_to_streaming_image_sse_impl() {
                 priority: 1,
                 api_formats: Some(vec!["openai:image".to_string()]),
                 endpoint_ids: None,
+                operations: None,
             }]),
             model_supports_streaming: Some(true),
             model_is_active: true,
@@ -840,6 +842,7 @@ fn image_bridge_candidate_row(
             priority: 1,
             api_formats: Some(vec!["openai:image".to_string()]),
             endpoint_ids: None,
+            operations: None,
         }]),
         model_supports_streaming: Some(false),
         model_is_active: true,
diff --git a/apps/aether-gateway/src/tests/ai_execute/stream/pii_redaction.rs b/apps/aether-gateway/src/tests/ai_execute/stream/pii_redaction.rs
index c6cfabde1..3f3b276dc 100644
--- a/apps/aether-gateway/src/tests/ai_execute/stream/pii_redaction.rs
+++ b/apps/aether-gateway/src/tests/ai_execute/stream/pii_redaction.rs
@@ -150,6 +150,7 @@ fn candidate_row() -> StoredMinimalCandidateSelectionRow {
             priority: 1,
             api_formats: Some(vec!["openai:chat".to_string()]),
             endpoint_ids: Some(vec!["endpoint-ai-execute-stream-pii-redaction".to_string()]),
+            operations: None,
         }]),
         model_supports_streaming: Some(true),
         model_is_active: true,
diff --git a/apps/aether-gateway/src/tests/ai_execute/stream_cli/compact.rs b/apps/aether-gateway/src/tests/ai_execute/stream_cli/compact.rs
index c7a0a8841..2d7207c5a 100644
--- a/apps/aether-gateway/src/tests/ai_execute/stream_cli/compact.rs
+++ b/apps/aether-gateway/src/tests/ai_execute/stream_cli/compact.rs
@@ -123,6 +123,7 @@ async fn gateway_executes_openai_responses_compact_as_unary_request_impl() {
                 priority: 1,
                 api_formats: Some(vec!["openai:responses:compact".to_string()]),
                 endpoint_ids: None,
+                operations: None,
             }]),
             model_supports_streaming: Some(true),
             model_is_active: true,
diff --git a/apps/aether-gateway/src/tests/ai_execute/stream_cli/direct.rs b/apps/aether-gateway/src/tests/ai_execute/stream_cli/direct.rs
index d885cc479..6842b23b9 100644
--- a/apps/aether-gateway/src/tests/ai_execute/stream_cli/direct.rs
+++ b/apps/aether-gateway/src/tests/ai_execute/stream_cli/direct.rs
@@ -136,6 +136,7 @@ async fn gateway_executes_codex_cli_stream_via_local_decision_gate_after_oauth_r
                 priority: 1,
                 api_formats: Some(vec!["openai:responses".to_string()]),
                 endpoint_ids: None,
+                operations: None,
             }]),
             model_supports_streaming: Some(true),
             model_is_active: true,
diff --git a/apps/aether-gateway/src/tests/ai_execute/stream_provider.rs b/apps/aether-gateway/src/tests/ai_execute/stream_provider.rs
index 06566661f..68ec79da7 100644
--- a/apps/aether-gateway/src/tests/ai_execute/stream_provider.rs
+++ b/apps/aether-gateway/src/tests/ai_execute/stream_provider.rs
@@ -182,6 +182,7 @@ async fn gateway_executes_kiro_claude_cli_stream_via_local_provider_catalog_cand
                 priority: 1,
                 api_formats: Some(vec!["claude:messages".to_string()]),
                 endpoint_ids: None,
+                operations: None,
             }]),
             model_supports_streaming: Some(true),
             model_is_active: true,
@@ -700,6 +701,7 @@ async fn gateway_executes_claude_cli_stream_via_local_decision_gate_without_wait
                 priority: 1,
                 api_formats: Some(vec!["claude:messages".to_string()]),
                 endpoint_ids: None,
+                operations: None,
             }]),
             model_supports_streaming: Some(true),
             model_is_active: true,
@@ -1161,6 +1163,7 @@ async fn gateway_executes_claude_code_cli_stream_via_local_decision_gate_with_lo
                 priority: 1,
                 api_formats: Some(vec!["claude:messages".to_string()]),
                 endpoint_ids: None,
+                operations: None,
             }]),
             model_supports_streaming: Some(true),
             model_is_active: true,
@@ -1694,6 +1697,7 @@ async fn gateway_executes_claude_chat_stream_via_local_decision_gate_with_local_
                 priority: 1,
                 api_formats: Some(vec!["claude:messages".to_string()]),
                 endpoint_ids: None,
+                operations: None,
             }]),
             model_supports_streaming: Some(true),
             model_is_active: true,
diff --git a/apps/aether-gateway/src/tests/ai_execute/stream_provider_gemini/local_chat.rs b/apps/aether-gateway/src/tests/ai_execute/stream_provider_gemini/local_chat.rs
index a3852bde8..3b2a21f91 100644
--- a/apps/aether-gateway/src/tests/ai_execute/stream_provider_gemini/local_chat.rs
+++ b/apps/aether-gateway/src/tests/ai_execute/stream_provider_gemini/local_chat.rs
@@ -101,6 +101,7 @@ async fn gateway_executes_gemini_chat_stream_via_local_decision_gate_with_local_
                 priority: 1,
                 api_formats: Some(vec!["gemini:generate_content".to_string()]),
                 endpoint_ids: None,
+                operations: None,
             }]),
             model_supports_streaming: Some(true),
             model_is_active: true,
diff --git a/apps/aether-gateway/src/tests/ai_execute/stream_provider_gemini/local_cli.rs b/apps/aether-gateway/src/tests/ai_execute/stream_provider_gemini/local_cli.rs
index 5e2b73e5e..7b7d98ad1 100644
--- a/apps/aether-gateway/src/tests/ai_execute/stream_provider_gemini/local_cli.rs
+++ b/apps/aether-gateway/src/tests/ai_execute/stream_provider_gemini/local_cli.rs
@@ -101,6 +101,7 @@ async fn gateway_executes_gemini_cli_stream_via_local_decision_gate_with_local_s
                 priority: 1,
                 api_formats: Some(vec!["gemini:generate_content".to_string()]),
                 endpoint_ids: None,
+                operations: None,
             }]),
             model_supports_streaming: Some(true),
             model_is_active: true,
@@ -553,6 +554,7 @@ async fn gateway_executes_gemini_cli_stream_via_local_decision_gate_after_oauth_
                 priority: 1,
                 api_formats: Some(vec!["gemini:generate_content".to_string()]),
                 endpoint_ids: None,
+                operations: None,
             }]),
             model_supports_streaming: Some(true),
             model_is_active: true,
@@ -1108,6 +1110,7 @@ async fn gateway_executes_vertex_ai_gemini_cli_stream_via_local_decision_gate_wi
                 priority: 1,
                 api_formats: Some(vec!["gemini:generate_content".to_string()]),
                 endpoint_ids: None,
+                operations: None,
             }]),
             model_supports_streaming: Some(true),
             model_is_active: true,
@@ -1573,6 +1576,7 @@ async fn gateway_executes_antigravity_gemini_cli_stream_via_local_decision_gate_
                 priority: 1,
                 api_formats: Some(vec!["gemini:generate_content".to_string()]),
                 endpoint_ids: None,
+                operations: None,
             }]),
             model_supports_streaming: Some(true),
             model_is_active: true,
diff --git a/apps/aether-gateway/src/tests/ai_execute/sync/chat/failover.rs b/apps/aether-gateway/src/tests/ai_execute/sync/chat/failover.rs
index eda10935b..cc8de1774 100644
--- a/apps/aether-gateway/src/tests/ai_execute/sync/chat/failover.rs
+++ b/apps/aether-gateway/src/tests/ai_execute/sync/chat/failover.rs
@@ -91,6 +91,7 @@ async fn gateway_skips_unsupported_local_openai_chat_sync_candidate_before_tryin
                 priority: 1,
                 api_formats: Some(vec!["openai:chat".to_string()]),
                 endpoint_ids: None,
+                operations: None,
             }]),
             model_supports_streaming: Some(true),
             model_is_active: true,
@@ -296,6 +297,7 @@ async fn gateway_skips_unsupported_local_openai_chat_sync_candidate_before_tryin
         priority: 1,
         api_formats: Some(vec!["openai:chat".to_string()]),
         endpoint_ids: None,
+        operations: None,
     }]);
     let candidate_selection_repository =
         Arc::new(InMemoryMinimalCandidateSelectionReadRepository::seed(vec![
@@ -497,6 +499,7 @@ async fn gateway_surfaces_local_execution_runtime_miss_reason_when_all_openai_ch
                 priority: 1,
                 api_formats: Some(vec!["openai:chat".to_string()]),
                 endpoint_ids: None,
+                operations: None,
             }]),
             model_supports_streaming: Some(true),
             model_is_active: true,
@@ -775,6 +778,7 @@ async fn gateway_retries_next_local_openai_chat_sync_candidate_after_auth_failur
                 priority: 1,
                 api_formats: Some(vec!["openai:chat".to_string()]),
                 endpoint_ids: None,
+                operations: None,
             }]),
             model_supports_streaming: Some(true),
             model_is_active: true,
diff --git a/apps/aether-gateway/src/tests/ai_execute/sync/chat/local_decision.rs b/apps/aether-gateway/src/tests/ai_execute/sync/chat/local_decision.rs
index 2bd34498a..9d92d1fd6 100644
--- a/apps/aether-gateway/src/tests/ai_execute/sync/chat/local_decision.rs
+++ b/apps/aether-gateway/src/tests/ai_execute/sync/chat/local_decision.rs
@@ -150,6 +150,7 @@ async fn proxy_pii_redaction_local_openai_chat_runtime_masks_headers_and_restore
                 priority: 1,
                 api_formats: Some(vec!["openai:chat".to_string()]),
                 endpoint_ids: Some(vec!["endpoint-redaction-1".to_string()]),
+                operations: None,
             }]),
             model_supports_streaming: Some(true),
             model_is_active: true,
@@ -454,6 +455,7 @@ async fn gateway_executes_openai_chat_sync_via_local_decision_gate_without_execu
                 priority: 1,
                 api_formats: Some(vec!["openai:chat".to_string()]),
                 endpoint_ids: None,
+                operations: None,
             }]),
             model_supports_streaming: Some(true),
             model_is_active: true,
@@ -647,6 +649,7 @@ async fn gateway_executes_openai_chat_sync_via_local_decision_gate_without_execu
         priority: 1,
         api_formats: Some(vec!["openai:chat".to_string()]),
         endpoint_ids: None,
+        operations: None,
     }]);
     let candidate_selection_repository =
         Arc::new(InMemoryMinimalCandidateSelectionReadRepository::seed(vec![
@@ -839,6 +842,7 @@ async fn gateway_executes_openai_chat_sync_with_regex_model_mapping_in_execution
                 priority: 1,
                 api_formats: Some(vec!["openai:chat".to_string()]),
                 endpoint_ids: None,
+                operations: None,
             }]),
             model_supports_streaming: Some(true),
             model_is_active: true,
@@ -1130,6 +1134,7 @@ async fn gateway_executes_openai_chat_sync_via_local_cross_format_gemini_candida
                 priority: 1,
                 api_formats: Some(vec!["gemini:generate_content".to_string()]),
                 endpoint_ids: None,
+                operations: None,
             }]),
             model_supports_streaming: Some(true),
             model_is_active: true,
@@ -1151,6 +1156,7 @@ async fn gateway_executes_openai_chat_sync_via_local_cross_format_gemini_candida
             priority: 2,
             api_formats: Some(vec!["gemini:generate_content".to_string()]),
             endpoint_ids: None,
+            operations: None,
         }]);
         row
     }
@@ -1698,6 +1704,7 @@ async fn gateway_returns_openai_chat_error_for_local_cross_format_claude_cli_syn
                 priority: 1,
                 api_formats: Some(vec!["claude:messages".to_string()]),
                 endpoint_ids: None,
+                operations: None,
             }]),
             model_supports_streaming: Some(true),
             model_is_active: true,
@@ -2102,6 +2109,7 @@ async fn gateway_returns_openai_chat_error_for_local_cross_format_gemini_cli_syn
                 priority: 1,
                 api_formats: Some(vec!["gemini:generate_content".to_string()]),
                 endpoint_ids: None,
+                operations: None,
             }]),
             model_supports_streaming: Some(true),
             model_is_active: true,
@@ -2531,6 +2539,7 @@ async fn gateway_returns_openai_chat_error_for_local_cross_format_claude_sync_fa
                 priority: 1,
                 api_formats: Some(vec!["claude:messages".to_string()]),
                 endpoint_ids: None,
+                operations: None,
             }]),
             model_supports_streaming: Some(true),
             model_is_active: true,
@@ -2939,6 +2948,7 @@ async fn gateway_returns_openai_chat_error_for_local_cross_format_gemini_sync_fa
                 priority: 1,
                 api_formats: Some(vec!["gemini:generate_content".to_string()]),
                 endpoint_ids: None,
+                operations: None,
             }]),
             model_supports_streaming: Some(true),
             model_is_active: true,
@@ -3387,6 +3397,7 @@ async fn gateway_executes_openai_chat_sync_with_custom_path_via_local_decision_g
                 priority: 1,
                 api_formats: Some(vec!["openai:chat".to_string()]),
                 endpoint_ids: None,
+                operations: None,
             }]),
             model_supports_streaming: Some(true),
             model_is_active: true,
diff --git a/apps/aether-gateway/src/tests/ai_execute/sync/chat/pii_redaction.rs b/apps/aether-gateway/src/tests/ai_execute/sync/chat/pii_redaction.rs
index 537f7c120..ad5f6866b 100644
--- a/apps/aether-gateway/src/tests/ai_execute/sync/chat/pii_redaction.rs
+++ b/apps/aether-gateway/src/tests/ai_execute/sync/chat/pii_redaction.rs
@@ -113,6 +113,7 @@ fn candidate_row(test_id: &str) -> StoredMinimalCandidateSelectionRow {
             priority: 1,
             api_formats: Some(vec!["openai:chat".to_string()]),
             endpoint_ids: Some(vec![format!("endpoint-{test_id}")]),
+            operations: None,
         }]),
         model_supports_streaming: Some(true),
         model_is_active: true,
diff --git a/apps/aether-gateway/src/tests/ai_execute/sync/claude/claude_code.rs b/apps/aether-gateway/src/tests/ai_execute/sync/claude/claude_code.rs
index c365a7d36..e5d38946a 100644
--- a/apps/aether-gateway/src/tests/ai_execute/sync/claude/claude_code.rs
+++ b/apps/aether-gateway/src/tests/ai_execute/sync/claude/claude_code.rs
@@ -129,6 +129,7 @@ async fn gateway_executes_claude_code_cli_sync_via_local_decision_gate_with_loca
                 priority: 1,
                 api_formats: Some(vec!["claude:messages".to_string()]),
                 endpoint_ids: None,
+                operations: None,
             }]),
             model_supports_streaming: Some(true),
             model_is_active: true,
diff --git a/apps/aether-gateway/src/tests/ai_execute/sync/claude/kiro.rs b/apps/aether-gateway/src/tests/ai_execute/sync/claude/kiro.rs
index 6239a7a1c..16a3db78a 100644
--- a/apps/aether-gateway/src/tests/ai_execute/sync/claude/kiro.rs
+++ b/apps/aether-gateway/src/tests/ai_execute/sync/claude/kiro.rs
@@ -197,6 +197,7 @@ async fn gateway_executes_kiro_claude_cli_sync_via_local_provider_catalog_candid
                 priority: 1,
                 api_formats: Some(vec!["claude:messages".to_string()]),
                 endpoint_ids: None,
+                operations: None,
             }]),
             model_supports_streaming: Some(true),
             model_is_active: true,
@@ -835,6 +836,7 @@ async fn gateway_executes_kiro_claude_cli_sync_via_local_provider_catalog_candid
                 priority: 1,
                 api_formats: Some(vec!["claude:messages".to_string()]),
                 endpoint_ids: None,
+                operations: None,
             }]),
             model_supports_streaming: Some(true),
             model_is_active: true,
diff --git a/apps/aether-gateway/src/tests/ai_execute/sync/claude/local_chat.rs b/apps/aether-gateway/src/tests/ai_execute/sync/claude/local_chat.rs
index f5f11be5b..733c60669 100644
--- a/apps/aether-gateway/src/tests/ai_execute/sync/claude/local_chat.rs
+++ b/apps/aether-gateway/src/tests/ai_execute/sync/claude/local_chat.rs
@@ -124,6 +124,7 @@ async fn gateway_executes_claude_chat_sync_via_local_decision_gate_with_local_sy
                 priority: 1,
                 api_formats: Some(vec!["claude:messages".to_string()]),
                 endpoint_ids: None,
+                operations: None,
             }]),
             model_supports_streaming: Some(true),
             model_is_active: true,
@@ -694,6 +695,7 @@ async fn gateway_returns_claude_chat_error_for_local_sync_failure_impl() {
                 priority: 1,
                 api_formats: Some(vec!["claude:messages".to_string()]),
                 endpoint_ids: None,
+                operations: None,
             }]),
             model_supports_streaming: Some(true),
             model_is_active: true,
diff --git a/apps/aether-gateway/src/tests/ai_execute/sync/claude/local_cli.rs b/apps/aether-gateway/src/tests/ai_execute/sync/claude/local_cli.rs
index eabb52d58..4687c1aa4 100644
--- a/apps/aether-gateway/src/tests/ai_execute/sync/claude/local_cli.rs
+++ b/apps/aether-gateway/src/tests/ai_execute/sync/claude/local_cli.rs
@@ -124,6 +124,7 @@ async fn gateway_executes_claude_cli_sync_via_local_decision_gate_with_local_syn
                 priority: 1,
                 api_formats: Some(vec!["claude:messages".to_string()]),
                 endpoint_ids: None,
+                operations: None,
             }]),
             model_supports_streaming: Some(true),
             model_is_active: true,
@@ -570,6 +571,7 @@ async fn gateway_returns_claude_cli_error_for_local_sync_failure_impl() {
                 priority: 1,
                 api_formats: Some(vec!["claude:messages".to_string()]),
                 endpoint_ids: None,
+                operations: None,
             }]),
             model_supports_streaming: Some(true),
             model_is_active: true,
@@ -853,6 +855,7 @@ async fn gateway_marks_claude_cli_cross_format_runtime_miss_when_format_conversi
                 priority: 1,
                 api_formats: Some(vec!["openai:responses".to_string()]),
                 endpoint_ids: None,
+                operations: None,
             }]),
             model_supports_streaming: Some(true),
             model_is_active: true,
diff --git a/apps/aether-gateway/src/tests/ai_execute/sync/cli.rs b/apps/aether-gateway/src/tests/ai_execute/sync/cli.rs
index 7c8f16999..b15c80c62 100644
--- a/apps/aether-gateway/src/tests/ai_execute/sync/cli.rs
+++ b/apps/aether-gateway/src/tests/ai_execute/sync/cli.rs
@@ -141,6 +141,7 @@ async fn gateway_executes_openai_responses_sync_via_local_decision_gate_with_loc
                 priority: 1,
                 api_formats: Some(vec!["openai:responses".to_string()]),
                 endpoint_ids: None,
+                operations: None,
             }]),
             model_supports_streaming: Some(true),
             model_is_active: true,
@@ -650,6 +651,7 @@ async fn gateway_waits_for_api_key_concurrency_slot_then_executes_openai_respons
                 priority: 1,
                 api_formats: Some(vec!["openai:responses".to_string()]),
                 endpoint_ids: None,
+                operations: None,
             }]),
             model_supports_streaming: Some(true),
             model_is_active: true,
@@ -1042,6 +1044,7 @@ async fn gateway_executes_openai_responses_sync_after_api_key_concurrency_wait_b
                 priority: 1,
                 api_formats: Some(vec!["openai:responses".to_string()]),
                 endpoint_ids: None,
+                operations: None,
             }]),
             model_supports_streaming: Some(true),
             model_is_active: true,
@@ -1412,6 +1415,7 @@ async fn gateway_returns_openai_responses_error_for_local_sync_failure_impl() {
                 priority: 1,
                 api_formats: Some(vec!["openai:responses".to_string()]),
                 endpoint_ids: None,
+                operations: None,
             }]),
             model_supports_streaming: Some(true),
             model_is_active: true,
@@ -1725,6 +1729,7 @@ async fn gateway_returns_openai_responses_error_for_local_cross_format_gemini_cl
                 priority: 1,
                 api_formats: Some(vec!["gemini:generate_content".to_string()]),
                 endpoint_ids: None,
+                operations: None,
             }]),
             model_supports_streaming: Some(true),
             model_is_active: true,
@@ -2131,6 +2136,7 @@ async fn gateway_returns_openai_responses_error_for_local_cross_format_claude_sy
                 priority: 1,
                 api_formats: Some(vec!["claude:messages".to_string()]),
                 endpoint_ids: None,
+                operations: None,
             }]),
             model_supports_streaming: Some(true),
             model_is_active: true,
@@ -2516,6 +2522,7 @@ async fn gateway_returns_openai_responses_error_for_local_cross_format_claude_ch
                 priority: 1,
                 api_formats: Some(vec!["claude:messages".to_string()]),
                 endpoint_ids: None,
+                operations: None,
             }]),
             model_supports_streaming: Some(true),
             model_is_active: true,
@@ -2904,6 +2911,7 @@ async fn gateway_returns_openai_responses_error_for_local_cross_format_gemini_ch
                 priority: 1,
                 api_formats: Some(vec!["gemini:generate_content".to_string()]),
                 endpoint_ids: None,
+                operations: None,
             }]),
             model_supports_streaming: Some(true),
             model_is_active: true,
@@ -3301,6 +3309,7 @@ async fn gateway_executes_codex_cli_sync_via_local_decision_gate_after_oauth_ref
                 priority: 1,
                 api_formats: Some(vec!["openai:responses".to_string()]),
                 endpoint_ids: None,
+                operations: None,
             }]),
             model_supports_streaming: Some(true),
             model_is_active: true,
diff --git a/apps/aether-gateway/src/tests/ai_execute/sync/gemini/cli.rs b/apps/aether-gateway/src/tests/ai_execute/sync/gemini/cli.rs
index 90034fe5c..770bd5846 100644
--- a/apps/aether-gateway/src/tests/ai_execute/sync/gemini/cli.rs
+++ b/apps/aether-gateway/src/tests/ai_execute/sync/gemini/cli.rs
@@ -123,6 +123,7 @@ async fn gateway_executes_gemini_cli_sync_via_local_decision_gate_with_local_syn
                 priority: 1,
                 api_formats: Some(vec!["gemini:generate_content".to_string()]),
                 endpoint_ids: None,
+                operations: None,
             }]),
             model_supports_streaming: Some(true),
             model_is_active: true,
@@ -561,6 +562,7 @@ async fn gateway_returns_gemini_cli_error_for_local_sync_failure_impl() {
                 priority: 1,
                 api_formats: Some(vec!["gemini:generate_content".to_string()]),
                 endpoint_ids: None,
+                operations: None,
             }]),
             model_supports_streaming: Some(true),
             model_is_active: true,
@@ -864,6 +866,7 @@ async fn gateway_executes_gemini_cli_sync_via_local_decision_gate_after_oauth_re
                 priority: 1,
                 api_formats: Some(vec!["gemini:generate_content".to_string()]),
                 endpoint_ids: None,
+                operations: None,
             }]),
             model_supports_streaming: Some(true),
             model_is_active: true,
@@ -1426,6 +1429,7 @@ async fn gateway_executes_vertex_ai_gemini_cli_sync_via_local_decision_gate_with
                 priority: 1,
                 api_formats: Some(vec!["gemini:generate_content".to_string()]),
                 endpoint_ids: None,
+                operations: None,
             }]),
             model_supports_streaming: Some(true),
             model_is_active: true,
@@ -1878,6 +1882,7 @@ async fn gateway_executes_antigravity_gemini_cli_sync_via_local_decision_gate_af
                 priority: 1,
                 api_formats: Some(vec!["gemini:generate_content".to_string()]),
                 endpoint_ids: None,
+                operations: None,
             }]),
             model_supports_streaming: Some(true),
             model_is_active: true,
diff --git a/apps/aether-gateway/src/tests/ai_execute/sync/gemini/local_chat.rs b/apps/aether-gateway/src/tests/ai_execute/sync/gemini/local_chat.rs
index 0e60c4ccf..79d89436e 100644
--- a/apps/aether-gateway/src/tests/ai_execute/sync/gemini/local_chat.rs
+++ b/apps/aether-gateway/src/tests/ai_execute/sync/gemini/local_chat.rs
@@ -123,6 +123,7 @@ async fn gateway_executes_gemini_chat_sync_via_local_decision_gate_with_local_sy
                 priority: 1,
                 api_formats: Some(vec!["gemini:generate_content".to_string()]),
                 endpoint_ids: None,
+                operations: None,
             }]),
             model_supports_streaming: Some(true),
             model_is_active: true,
@@ -551,6 +552,7 @@ async fn gateway_returns_gemini_chat_error_for_local_sync_failure_impl() {
                 priority: 1,
                 api_formats: Some(vec!["gemini:generate_content".to_string()]),
                 endpoint_ids: None,
+                operations: None,
             }]),
             model_supports_streaming: Some(true),
             model_is_active: true,
diff --git a/apps/aether-gateway/src/tests/ai_execute/sync/image.rs b/apps/aether-gateway/src/tests/ai_execute/sync/image.rs
index 915de228e..cd65a8a9b 100644
--- a/apps/aether-gateway/src/tests/ai_execute/sync/image.rs
+++ b/apps/aether-gateway/src/tests/ai_execute/sync/image.rs
@@ -129,6 +129,7 @@ async fn gateway_converts_openai_image_sync_to_gemini_image_provider_impl() {
                 priority: 1,
                 api_formats: Some(vec!["gemini:generate_content".to_string()]),
                 endpoint_ids: None,
+                operations: None,
             }]),
             model_supports_streaming: Some(true),
             model_is_active: true,
@@ -481,6 +482,7 @@ async fn gateway_converts_gemini_image_sync_to_openai_image_provider_impl() {
                 priority: 1,
                 api_formats: Some(vec!["openai:image".to_string()]),
                 endpoint_ids: None,
+                operations: None,
             }]),
             model_supports_streaming: Some(true),
             model_is_active: true,
@@ -833,6 +835,7 @@ async fn gateway_executes_codex_image_sync_via_local_decision_gate_after_oauth_r
                 priority: 1,
                 api_formats: Some(vec!["openai:image".to_string()]),
                 endpoint_ids: None,
+                operations: None,
             }]),
             model_supports_streaming: Some(true),
             model_is_active: true,
@@ -1242,6 +1245,7 @@ async fn gateway_plans_chatgpt_web_image_sync_with_internal_web_executor_url_imp
                 priority: 1,
                 api_formats: Some(vec!["openai:image".to_string()]),
                 endpoint_ids: None,
+                operations: None,
             }]),
             model_supports_streaming: Some(true),
             model_is_active: true,
diff --git a/apps/aether-gateway/src/tests/ai_execute/sync/pii_redaction_formats.rs b/apps/aether-gateway/src/tests/ai_execute/sync/pii_redaction_formats.rs
index 5137d1dc3..45e786726 100644
--- a/apps/aether-gateway/src/tests/ai_execute/sync/pii_redaction_formats.rs
+++ b/apps/aether-gateway/src/tests/ai_execute/sync/pii_redaction_formats.rs
@@ -608,6 +608,7 @@ fn candidate_row(case: &RedactionFormatCase) -> StoredMinimalCandidateSelectionR
             priority: 1,
             api_formats: Some(vec![case.provider_format.api_format().to_string()]),
             endpoint_ids: Some(vec![format!("endpoint-{}", case.test_id)]),
+            operations: None,
         }]),
         model_supports_streaming: Some(true),
         model_is_active: true,
diff --git a/apps/aether-gateway/src/tests/ai_execute/sync/search.rs b/apps/aether-gateway/src/tests/ai_execute/sync/search.rs
index eb1de4c43..fd11355bd 100644
--- a/apps/aether-gateway/src/tests/ai_execute/sync/search.rs
+++ b/apps/aether-gateway/src/tests/ai_execute/sync/search.rs
@@ -119,6 +119,7 @@ async fn gateway_executes_codex_search_with_responses_permission_and_search_cont
                 priority: 1,
                 api_formats: Some(vec!["openai:responses".to_string()]),
                 endpoint_ids: None,
+                operations: None,
             }]),
             model_supports_streaming: Some(false),
             model_is_active: true,
diff --git a/apps/aether-gateway/src/tests/audit.rs b/apps/aether-gateway/src/tests/audit.rs
index 497a329c9..776a0203e 100644
--- a/apps/aether-gateway/src/tests/audit.rs
+++ b/apps/aether-gateway/src/tests/audit.rs
@@ -112,6 +112,7 @@ fn sample_local_openai_candidate_row() -> StoredMinimalCandidateSelectionRow {
             priority: 1,
             api_formats: Some(vec!["openai:chat".to_string()]),
             endpoint_ids: None,
+            operations: None,
         }]),
         model_supports_streaming: Some(true),
         model_is_active: true,
diff --git a/apps/aether-gateway/src/tests/files/mod.rs b/apps/aether-gateway/src/tests/files/mod.rs
index 5c85a232a..bc374ba78 100644
--- a/apps/aether-gateway/src/tests/files/mod.rs
+++ b/apps/aether-gateway/src/tests/files/mod.rs
@@ -115,6 +115,7 @@ fn sample_files_candidate_row() -> StoredMinimalCandidateSelectionRow {
             priority: 1,
             api_formats: Some(vec!["gemini:files".to_string()]),
             endpoint_ids: None,
+            operations: None,
         }]),
         model_supports_streaming: Some(true),
         model_is_active: true,
diff --git a/apps/aether-gateway/src/tests/frontdoor.rs b/apps/aether-gateway/src/tests/frontdoor.rs
index 11a59093e..15ae3a645 100644
--- a/apps/aether-gateway/src/tests/frontdoor.rs
+++ b/apps/aether-gateway/src/tests/frontdoor.rs
@@ -291,6 +291,7 @@ fn sample_models_candidate_row(
             priority: 1,
             api_formats: Some(vec![api_format.to_string()]),
             endpoint_ids: None,
+            operations: None,
         }]),
         model_supports_streaming: Some(true),
         model_is_active: true,
diff --git a/apps/aether-gateway/src/tests/frontdoor/ai.rs b/apps/aether-gateway/src/tests/frontdoor/ai.rs
index 7e4951a61..2ddf5b6ab 100644
--- a/apps/aether-gateway/src/tests/frontdoor/ai.rs
+++ b/apps/aether-gateway/src/tests/frontdoor/ai.rs
@@ -74,6 +74,7 @@ fn sample_codex_models_candidate_row(
             priority: 1,
             api_formats: Some(vec!["openai:responses".to_string()]),
             endpoint_ids: None,
+            operations: None,
         },
     ]);
     row
diff --git a/apps/aether-gateway/src/tests/usage.rs b/apps/aether-gateway/src/tests/usage.rs
index b5c78b0a8..77e865e61 100644
--- a/apps/aether-gateway/src/tests/usage.rs
+++ b/apps/aether-gateway/src/tests/usage.rs
@@ -101,6 +101,7 @@ pub(super) fn sample_local_openai_candidate_row() -> StoredMinimalCandidateSelec
             priority: 1,
             api_formats: Some(vec!["openai:chat".to_string()]),
             endpoint_ids: None,
+            operations: None,
         }]),
         model_supports_streaming: Some(true),
         model_is_active: true,
diff --git a/apps/aether-gateway/src/tests/usage/local.rs b/apps/aether-gateway/src/tests/usage/local.rs
index 37135c491..ea8a63ec4 100644
--- a/apps/aether-gateway/src/tests/usage/local.rs
+++ b/apps/aether-gateway/src/tests/usage/local.rs
@@ -1611,6 +1611,7 @@ async fn gateway_records_failed_usage_when_all_local_claude_cli_candidates_are_s
                 priority: 1,
                 api_formats: Some(vec!["openai:responses".to_string()]),
                 endpoint_ids: None,
+                operations: None,
             }]),
             model_supports_streaming: Some(true),
             model_is_active: true,
@@ -1926,6 +1927,7 @@ fn gateway_keeps_failed_usage_request_capture_lightweight_for_large_local_claude
                 priority: 1,
                 api_formats: Some(vec!["openai:responses".to_string()]),
                 endpoint_ids: None,
+                operations: None,
             }]),
             model_supports_streaming: Some(true),
             model_is_active: true,
diff --git a/apps/aether-gateway/src/tests/usage/pricing.rs b/apps/aether-gateway/src/tests/usage/pricing.rs
index d277521a3..6a6986599 100644
--- a/apps/aether-gateway/src/tests/usage/pricing.rs
+++ b/apps/aether-gateway/src/tests/usage/pricing.rs
@@ -268,6 +268,7 @@ fn sample_candidate_row(spec: ProviderSpec) -> StoredMinimalCandidateSelectionRo
             priority: 1,
             api_formats: Some(vec![spec.api_format.to_string()]),
             endpoint_ids: None,
+            operations: None,
         }]),
         model_supports_streaming: Some(true),
         model_is_active: true,
diff --git a/apps/aether-gateway/src/tests/video/gemini_sync_create.rs b/apps/aether-gateway/src/tests/video/gemini_sync_create.rs
index ae96d6242..3a6db5bd8 100644
--- a/apps/aether-gateway/src/tests/video/gemini_sync_create.rs
+++ b/apps/aether-gateway/src/tests/video/gemini_sync_create.rs
@@ -110,6 +110,7 @@ async fn gateway_executes_gemini_video_create_via_local_decision_gate_with_local
                 priority: 1,
                 api_formats: Some(vec!["gemini:video".to_string()]),
                 endpoint_ids: None,
+                operations: None,
             }]),
             model_supports_streaming: Some(false),
             model_is_active: true,
diff --git a/apps/aether-gateway/src/tests/video/openai_sync_create.rs b/apps/aether-gateway/src/tests/video/openai_sync_create.rs
index a15e0ef1a..536c1939a 100644
--- a/apps/aether-gateway/src/tests/video/openai_sync_create.rs
+++ b/apps/aether-gateway/src/tests/video/openai_sync_create.rs
@@ -115,6 +115,7 @@ async fn gateway_executes_openai_video_create_via_local_decision_gate_with_local
                 priority: 1,
                 api_formats: Some(vec!["openai:video".to_string()]),
                 endpoint_ids: None,
+                operations: None,
             }]),
             model_supports_streaming: Some(false),
             model_is_active: true,
diff --git a/crates/aether-admin/src/observability/usage.rs b/crates/aether-admin/src/observability/usage.rs
index 5ac2f2cb0..36ff451c5 100644
--- a/crates/aether-admin/src/observability/usage.rs
+++ b/crates/aether-admin/src/observability/usage.rs
@@ -1188,6 +1188,7 @@ fn admin_usage_active_request_json(
     let mut value = json!({
         "id": item.id,
         "status": item.status,
+        "request_type": item.request_type,
         "input_tokens": item.input_tokens,
         "effective_input_tokens": admin_usage_effective_input_tokens(item),
         "output_tokens": item.output_tokens,
@@ -1304,6 +1305,7 @@ pub fn admin_usage_record_json(
         "status_code": item.status_code,
         "error_message": item.error_message,
         "status": item.status,
+        "request_type": item.request_type,
         "has_fallback": admin_usage_has_fallback(item),
         "has_retry": false,
         "has_rectified": false,
diff --git a/crates/aether-ai-formats/src/formats/openai/responses/mod.rs b/crates/aether-ai-formats/src/formats/openai/responses/mod.rs
index 16e73a0c7..7fca6658e 100644
--- a/crates/aether-ai-formats/src/formats/openai/responses/mod.rs
+++ b/crates/aether-ai-formats/src/formats/openai/responses/mod.rs
@@ -8,6 +8,34 @@ pub mod stream;
 
 const TOOL_ERROR_PREFIX: &str = "[tool error]";
 
+/// Semantic operation carried by an OpenAI Responses request that asks the
+/// service to compact a thread. The request still uses the Responses wire
+/// contract and transport endpoint.
+pub const OPENAI_RESPONSES_OPERATION_COMPACT: &str = "compact";
+
+/// Resolves the operation expressed by an OpenAI Responses wire request.
+///
+/// `responses_compaction_v2` is represented by a `compaction_trigger` input
+/// item on the normal Responses request. The legacy Compact API format is
+/// retained as the same operation for observability and scoped model mapping.
+pub fn openai_responses_request_operation(api_format: &str, body: &Value) -> Option<&'static str> {
+    if aether_ai_formats::is_openai_responses_compact_format(api_format) {
+        return Some(OPENAI_RESPONSES_OPERATION_COMPACT);
+    }
+    if !aether_ai_formats::is_openai_responses_format(api_format) {
+        return None;
+    }
+
+    body.get("input")
+        .and_then(Value::as_array)
+        .is_some_and(|items| {
+            items
+                .iter()
+                .any(|item| item.get("type").and_then(Value::as_str) == Some("compaction_trigger"))
+        })
+        .then_some(OPENAI_RESPONSES_OPERATION_COMPACT)
+}
+
 fn encode_tool_result_error(output: Value, is_error: bool) -> Value {
     if !is_error {
         return output;
@@ -23,3 +51,41 @@ fn encode_tool_result_error(output: Value, is_error: bool) -> Value {
         Value::String(format!("{TOOL_ERROR_PREFIX}\n{detail}"))
     }
 }
+
+#[cfg(test)]
+mod tests {
+    use serde_json::json;
+
+    use super::{openai_responses_request_operation, OPENAI_RESPONSES_OPERATION_COMPACT};
+
+    #[test]
+    fn resolves_compaction_trigger_as_compact_operation_on_responses_transport() {
+        assert_eq!(
+            openai_responses_request_operation(
+                "openai:responses",
+                &json!({
+                    "input": [
+                        {"role": "user", "content": "keep working"},
+                        {"type": "compaction_trigger"}
+                    ]
+                }),
+            ),
+            Some(OPENAI_RESPONSES_OPERATION_COMPACT)
+        );
+        assert_eq!(
+            openai_responses_request_operation(
+                "openai:responses",
+                &json!({"input": [{"role": "user", "content": "keep working"}]}),
+            ),
+            None
+        );
+    }
+
+    #[test]
+    fn resolves_legacy_compact_contract_without_a_body_marker() {
+        assert_eq!(
+            openai_responses_request_operation("openai:responses:compact", &json!({})),
+            Some(OPENAI_RESPONSES_OPERATION_COMPACT)
+        );
+    }
+}
diff --git a/crates/aether-ai-formats/src/lib.rs b/crates/aether-ai-formats/src/lib.rs
index b5ce22cf9..892943480 100644
--- a/crates/aether-ai-formats/src/lib.rs
+++ b/crates/aether-ai-formats/src/lib.rs
@@ -49,6 +49,9 @@ pub use formats::openai::responses::codex::{
 pub use formats::openai::responses::request::{
     validate_openai_responses_request_contract, OpenAiResponsesRequestContractViolation,
 };
+pub use formats::openai::responses::{
+    openai_responses_request_operation, OPENAI_RESPONSES_OPERATION_COMPACT,
+};
 pub use formats::registry::{
     build_stream_transcoder, convert_request, convert_request_pure,
     convert_request_pure_with_context, convert_response, convert_response_pure, emit_request_pure,
diff --git a/crates/aether-data-contracts/src/repository/candidate_selection/types.rs b/crates/aether-data-contracts/src/repository/candidate_selection/types.rs
index 309773285..46f160b6c 100644
--- a/crates/aether-data-contracts/src/repository/candidate_selection/types.rs
+++ b/crates/aether-data-contracts/src/repository/candidate_selection/types.rs
@@ -7,6 +7,10 @@ pub struct StoredProviderModelMapping {
     pub api_formats: Option<Vec<String>>,
     #[serde(default, skip_serializing_if = "Option::is_none")]
     pub endpoint_ids: Option<Vec<String>>,
+    /// Optional request-operation scope. An omitted scope applies to every
+    /// operation supported by the selected API format.
+    #[serde(default, skip_serializing_if = "Option::is_none")]
+    pub operations: Option<Vec<String>>,
 }
 
 #[derive(Debug, Clone, PartialEq, serde::Serialize, serde::Deserialize)]
diff --git a/crates/aether-data/src/repository/candidate_selection/memory.rs b/crates/aether-data/src/repository/candidate_selection/memory.rs
index 4b136bba4..25b811606 100644
--- a/crates/aether-data/src/repository/candidate_selection/memory.rs
+++ b/crates/aether-data/src/repository/candidate_selection/memory.rs
@@ -426,6 +426,7 @@ mod tests {
             priority: 0,
             api_formats: Some(vec!["openai:chat".to_string()]),
             endpoint_ids: None,
+            operations: None,
         }]);
         let repository = InMemoryMinimalCandidateSelectionReadRepository::seed(vec![
             mapped,
@@ -458,6 +459,7 @@ mod tests {
             priority: 0,
             api_formats: Some(vec!["openai:responses".to_string()]),
             endpoint_ids: None,
+            operations: None,
         }]);
 
         let mut responses = search.clone();
@@ -521,6 +523,7 @@ mod tests {
             priority: 1,
             api_formats: None,
             endpoint_ids: Some(vec!["endpoint-openai".to_string()]),
+            operations: None,
         }]);
 
         let mut scoped_out = selected.clone();
diff --git a/crates/aether-data/src/repository/candidate_selection/mysql.rs b/crates/aether-data/src/repository/candidate_selection/mysql.rs
index 9ae01cb74..5fe1eedcb 100644
--- a/crates/aether-data/src/repository/candidate_selection/mysql.rs
+++ b/crates/aether-data/src/repository/candidate_selection/mysql.rs
@@ -677,6 +677,7 @@ fn parse_embedded_provider_model_mappings(
         priority: 1,
         api_formats: None,
         endpoint_ids: None,
+        operations: None,
     }]))
 }
 
@@ -697,6 +698,7 @@ fn parse_provider_model_mappings_array(
                     priority: 1,
                     api_formats: None,
                     endpoint_ids: None,
+                    operations: None,
                 });
             }
             _ => {}
@@ -741,6 +743,11 @@ fn parse_provider_model_mapping_object_lenient(
         object.get("endpoint_ids").cloned(),
         "models.provider_model_mappings.endpoint_ids",
     )?;
+    let operations = parse_string_list(
+        object.get("operations").cloned(),
+        "models.provider_model_mappings.operations",
+    )?
+    .and_then(normalize_request_operations);
 
     Ok(Some(StoredProviderModelMapping {
         name: name.to_string(),
@@ -751,9 +758,19 @@ fn parse_provider_model_mapping_object_lenient(
         })?,
         api_formats,
         endpoint_ids,
+        operations,
     }))
 }
 
+fn normalize_request_operations(values: Vec<String>) -> Option<Vec<String>> {
+    let operations = values
+        .into_iter()
+        .map(|value| value.trim().to_ascii_lowercase())
+        .filter(|value| !value.is_empty())
+        .collect::<Vec<_>>();
+    (!operations.is_empty()).then_some(operations)
+}
+
 fn api_format_aliases(api_format: &str) -> Vec<String> {
     aether_ai_formats::api_format_storage_aliases(api_format)
 }
diff --git a/crates/aether-data/src/repository/candidate_selection/postgres.rs b/crates/aether-data/src/repository/candidate_selection/postgres.rs
index ccf7d5835..2759d119c 100644
--- a/crates/aether-data/src/repository/candidate_selection/postgres.rs
+++ b/crates/aether-data/src/repository/candidate_selection/postgres.rs
@@ -1350,6 +1350,7 @@ fn parse_embedded_provider_model_mappings(
         priority: 1,
         api_formats: None,
         endpoint_ids: None,
+        operations: None,
     }]))
 }
 
@@ -1372,6 +1373,7 @@ fn parse_provider_model_mappings_array(
                         priority: 1,
                         api_formats: None,
                         endpoint_ids: None,
+                        operations: None,
                     });
                 }
             }
@@ -1428,6 +1430,11 @@ fn parse_provider_model_mapping_object_lenient(
         object.get("endpoint_ids").cloned(),
         "models.provider_model_mappings.endpoint_ids",
     )?;
+    let operations = parse_string_list(
+        object.get("operations").cloned(),
+        "models.provider_model_mappings.operations",
+    )?
+    .and_then(normalize_request_operations);
 
     Ok(Some(StoredProviderModelMapping {
         name: name.to_string(),
@@ -1438,9 +1445,19 @@ fn parse_provider_model_mapping_object_lenient(
         })?,
         api_formats,
         endpoint_ids,
+        operations,
     }))
 }
 
+fn normalize_request_operations(values: Vec<String>) -> Option<Vec<String>> {
+    let operations = values
+        .into_iter()
+        .map(|value| value.trim().to_ascii_lowercase())
+        .filter(|value| !value.is_empty())
+        .collect::<Vec<_>>();
+    (!operations.is_empty()).then_some(operations)
+}
+
 #[cfg(test)]
 mod tests {
     use serde_json::json;
@@ -1624,7 +1641,7 @@ mod tests {
     #[test]
     fn parse_provider_model_mappings_accepts_stringified_array() {
         let parsed = parse_provider_model_mappings(Some(json!(
-            "[{\"name\":\"gpt-5.2\",\"priority\":2,\"api_formats\":[\"openai:chat\"]}]"
+            "[{\"name\":\"gpt-5.2\",\"priority\":2,\"api_formats\":[\"openai:chat\"],\"operations\":[\"COMPACT\"]}]"
         )))
         .expect("stringified provider_model_mappings should parse");
 
@@ -1635,6 +1652,7 @@ mod tests {
                 priority: 2,
                 api_formats: Some(vec!["openai:chat".to_string()]),
                 endpoint_ids: None,
+                operations: Some(vec!["compact".to_string()]),
             }])
         );
     }
@@ -1651,6 +1669,7 @@ mod tests {
                 priority: 1,
                 api_formats: None,
                 endpoint_ids: None,
+                operations: None,
             }])
         );
     }
@@ -1674,12 +1693,14 @@ mod tests {
                     priority: 1,
                     api_formats: None,
                     endpoint_ids: None,
+                    operations: None,
                 },
                 StoredProviderModelMapping {
                     name: "gpt-5.2-mini".to_string(),
                     priority: 1,
                     api_formats: None,
                     endpoint_ids: None,
+                    operations: None,
                 }
             ])
         );
diff --git a/crates/aether-data/src/repository/candidate_selection/sqlite.rs b/crates/aether-data/src/repository/candidate_selection/sqlite.rs
index 4190d56af..4fda5f725 100644
--- a/crates/aether-data/src/repository/candidate_selection/sqlite.rs
+++ b/crates/aether-data/src/repository/candidate_selection/sqlite.rs
@@ -1070,6 +1070,7 @@ fn parse_embedded_provider_model_mappings(
         priority: 1,
         api_formats: None,
         endpoint_ids: None,
+        operations: None,
     }]))
 }
 
@@ -1090,6 +1091,7 @@ fn parse_provider_model_mappings_array(
                     priority: 1,
                     api_formats: None,
                     endpoint_ids: None,
+                    operations: None,
                 });
             }
             _ => {}
@@ -1134,6 +1136,11 @@ fn parse_provider_model_mapping_object_lenient(
         object.get("endpoint_ids").cloned(),
         "models.provider_model_mappings.endpoint_ids",
     )?;
+    let operations = parse_string_list(
+        object.get("operations").cloned(),
+        "models.provider_model_mappings.operations",
+    )?
+    .and_then(normalize_request_operations);
 
     Ok(Some(StoredProviderModelMapping {
         name: name.to_string(),
@@ -1144,9 +1151,19 @@ fn parse_provider_model_mapping_object_lenient(
         })?,
         api_formats,
         endpoint_ids,
+        operations,
     }))
 }
 
+fn normalize_request_operations(values: Vec<String>) -> Option<Vec<String>> {
+    let operations = values
+        .into_iter()
+        .map(|value| value.trim().to_ascii_lowercase())
+        .filter(|value| !value.is_empty())
+        .collect::<Vec<_>>();
+    (!operations.is_empty()).then_some(operations)
+}
+
 fn api_format_aliases(api_format: &str) -> Vec<String> {
     aether_ai_formats::api_format_storage_aliases(api_format)
 }
@@ -1211,6 +1228,14 @@ mod tests {
             Some(vec!["alias-global".to_string()])
         );
         assert_eq!(rows[1].global_model_supports_streaming, Some(true));
+        assert_eq!(
+            rows[1]
+                .model_provider_model_mappings
+                .as_ref()
+                .and_then(|mappings| mappings.first())
+                .and_then(|mapping| mapping.operations.as_ref()),
+            Some(&vec!["compact".to_string()])
+        );
 
         let requested = repository
             .list_for_exact_api_format_and_requested_model_page(
@@ -1392,7 +1417,7 @@ INSERT INTO models (
 )
 VALUES (
   'model-1', 'provider-1', 'global-1', 'provider-model',
-  '[{"name":"alias-provider","api_formats":["openai:chat"],"priority":1}]',
+  '[{"name":"alias-provider","api_formats":["openai:chat"],"operations":["COMPACT"],"priority":1}]',
   1, 1, 1, 1, 1
 ),
 (
diff --git a/crates/aether-scheduler-core/src/candidate/enumeration.rs b/crates/aether-scheduler-core/src/candidate/enumeration.rs
index 07472e545..892bed4aa 100644
--- a/crates/aether-scheduler-core/src/candidate/enumeration.rs
+++ b/crates/aether-scheduler-core/src/candidate/enumeration.rs
@@ -27,6 +27,7 @@ fn enumerate_minimal_candidate_selection_inner(
     let EnumerateMinimalCandidateSelectionInput {
         rows,
         normalized_api_format,
+        request_operation,
         requested_model_name,
         resolved_global_model_name,
         require_streaming,
@@ -63,11 +64,12 @@ fn enumerate_minimal_candidate_selection_inner(
             continue;
         }
         let Some((selected_provider_model_name, mapping_matched_model)) =
-            crate::resolve_provider_model_name_with_model_directives(
+            crate::resolve_provider_model_name_with_model_directives_and_request_operation(
                 &row,
                 requested_model_name,
                 normalized_api_format,
                 enable_model_directives,
+                request_operation,
             )
         else {
             continue;
diff --git a/crates/aether-scheduler-core/src/candidate/mod.rs b/crates/aether-scheduler-core/src/candidate/mod.rs
index 579102284..d27642b5b 100644
--- a/crates/aether-scheduler-core/src/candidate/mod.rs
+++ b/crates/aether-scheduler-core/src/candidate/mod.rs
@@ -71,6 +71,7 @@ mod tests {
                 priority: 1,
                 api_formats: Some(vec!["openai:chat".to_string()]),
                 endpoint_ids: None,
+                operations: None,
             }]),
             model_supports_streaming: None,
             model_is_active: true,
@@ -199,6 +200,7 @@ mod tests {
             super::enumerate_minimal_candidate_selection(EnumerateMinimalCandidateSelectionInput {
                 rows: vec![sample_row("1"), disallowed],
                 normalized_api_format: "openai:chat",
+                request_operation: None,
                 requested_model_name: "gpt-5",
                 resolved_global_model_name: "gpt-5",
                 require_streaming: false,
@@ -221,6 +223,7 @@ mod tests {
             super::enumerate_minimal_candidate_selection(EnumerateMinimalCandidateSelectionInput {
                 rows: vec![row],
                 normalized_api_format: "openai:chat",
+                request_operation: None,
                 requested_model_name: "gpt-5",
                 resolved_global_model_name: "gpt-5",
                 require_streaming: false,
@@ -244,6 +247,7 @@ mod tests {
             super::enumerate_minimal_candidate_selection(EnumerateMinimalCandidateSelectionInput {
                 rows: vec![later_priority, earlier_priority],
                 normalized_api_format: "openai:chat",
+                request_operation: None,
                 requested_model_name: "gpt-5",
                 resolved_global_model_name: "gpt-5",
                 require_streaming: false,
@@ -295,6 +299,7 @@ mod tests {
             super::enumerate_minimal_candidate_selection(EnumerateMinimalCandidateSelectionInput {
                 rows: vec![missing_capability, matching_capability],
                 normalized_api_format: "openai:chat",
+                request_operation: None,
                 requested_model_name: "gpt-5",
                 resolved_global_model_name: "gpt-5",
                 require_streaming: false,
diff --git a/crates/aether-scheduler-core/src/candidate/types.rs b/crates/aether-scheduler-core/src/candidate/types.rs
index dc6191e08..7aaff708b 100644
--- a/crates/aether-scheduler-core/src/candidate/types.rs
+++ b/crates/aether-scheduler-core/src/candidate/types.rs
@@ -32,6 +32,7 @@ pub struct SchedulerMinimalCandidateSelectionCandidate {
 pub struct EnumerateMinimalCandidateSelectionInput<'a> {
     pub rows: Vec<StoredMinimalCandidateSelectionRow>,
     pub normalized_api_format: &'a str,
+    pub request_operation: Option<&'a str>,
     pub requested_model_name: &'a str,
     pub resolved_global_model_name: &'a str,
     pub require_streaming: bool,
diff --git a/crates/aether-scheduler-core/src/lib.rs b/crates/aether-scheduler-core/src/lib.rs
index ae6b26159..3fbfe0498 100644
--- a/crates/aether-scheduler-core/src/lib.rs
+++ b/crates/aether-scheduler-core/src/lib.rs
@@ -40,10 +40,13 @@ pub use health::{
 pub use model::{
     candidate_model_names, extract_global_priority_for_format, matches_model_mapping,
     normalize_api_format, resolve_provider_model_name,
-    resolve_provider_model_name_with_model_directives, resolve_requested_global_model_name,
-    resolve_requested_global_model_name_with_model_directives, row_supports_requested_model,
-    row_supports_requested_model_with_model_directives, row_supports_required_capability,
-    select_provider_model_name,
+    resolve_provider_model_name_with_model_directives,
+    resolve_provider_model_name_with_model_directives_and_request_operation,
+    resolve_requested_global_model_name, resolve_requested_global_model_name_with_model_directives,
+    resolve_requested_global_model_name_with_model_directives_and_request_operation,
+    row_supports_requested_model, row_supports_requested_model_with_model_directives,
+    row_supports_requested_model_with_model_directives_and_request_operation,
+    row_supports_required_capability, select_provider_model_name,
 };
 pub use provider::{build_provider_concurrent_limit_map, should_skip_provider_quota};
 pub use ranking::{
diff --git a/crates/aether-scheduler-core/src/model.rs b/crates/aether-scheduler-core/src/model.rs
index f143645f2..90f2f4bbc 100644
--- a/crates/aether-scheduler-core/src/model.rs
+++ b/crates/aether-scheduler-core/src/model.rs
@@ -25,17 +25,33 @@ pub fn resolve_requested_global_model_name_with_model_directives(
     requested_model_name: &str,
     api_format: &str,
     enable_model_directives: bool,
+) -> Option<String> {
+    resolve_requested_global_model_name_with_model_directives_and_request_operation(
+        rows,
+        requested_model_name,
+        api_format,
+        enable_model_directives,
+        None,
+    )
+}
+
+pub fn resolve_requested_global_model_name_with_model_directives_and_request_operation(
+    rows: &[StoredMinimalCandidateSelectionRow],
+    requested_model_name: &str,
+    api_format: &str,
+    enable_model_directives: bool,
+    request_operation: Option<&str>,
 ) -> Option<String> {
     requested_model_name_candidates(requested_model_name, enable_model_directives).find_map(
         |requested_model_name| {
             let requested_model_name = requested_model_name.as_ref();
             resolve_global_model_name_by(rows, |row| {
-                row_has_available_provider_model(row, api_format)
+                row_has_available_provider_model(row, api_format, request_operation)
                     && row.global_model_name == requested_model_name
             })
             .or_else(|| {
                 resolve_global_model_name_by(rows, |row| {
-                    row_default_provider_model_name_available(row, api_format)
+                    row_default_provider_model_name_available(row, api_format, request_operation)
                         && row.model_provider_model_name == requested_model_name
                 })
             })
@@ -45,7 +61,7 @@ pub fn resolve_requested_global_model_name_with_model_directives(
                         .as_ref()
                         .is_some_and(|mappings| {
                             mappings.iter().any(|mapping| {
-                                mapping_scope_matches(mapping, row, api_format)
+                                mapping_scope_matches(mapping, row, api_format, request_operation)
                                     && mapping.name == requested_model_name
                             })
                         })
@@ -53,7 +69,7 @@ pub fn resolve_requested_global_model_name_with_model_directives(
             })
             .or_else(|| {
                 resolve_global_model_name_by(rows, |row| {
-                    row_has_available_provider_model(row, api_format)
+                    row_has_available_provider_model(row, api_format, request_operation)
                         && row.global_model_mappings.as_ref().is_some_and(|patterns| {
                             patterns
                                 .iter()
@@ -78,10 +94,31 @@ pub fn row_supports_requested_model_with_model_directives(
     requested_model_name: &str,
     api_format: &str,
     enable_model_directives: bool,
+) -> bool {
+    row_supports_requested_model_with_model_directives_and_request_operation(
+        row,
+        requested_model_name,
+        api_format,
+        enable_model_directives,
+        None,
+    )
+}
+
+pub fn row_supports_requested_model_with_model_directives_and_request_operation(
+    row: &StoredMinimalCandidateSelectionRow,
+    requested_model_name: &str,
+    api_format: &str,
+    enable_model_directives: bool,
+    request_operation: Option<&str>,
 ) -> bool {
     requested_model_name_candidates(requested_model_name, enable_model_directives).any(
         |requested_model_name| {
-            row_supports_requested_model_exact(row, requested_model_name.as_ref(), api_format)
+            row_supports_requested_model_exact(
+                row,
+                requested_model_name.as_ref(),
+                api_format,
+                request_operation,
+            )
         },
     )
 }
@@ -90,10 +127,11 @@ fn row_supports_requested_model_exact(
     row: &StoredMinimalCandidateSelectionRow,
     requested_model_name: &str,
     api_format: &str,
+    request_operation: Option<&str>,
 ) -> bool {
-    row_has_available_provider_model(row, api_format)
+    row_has_available_provider_model(row, api_format, request_operation)
         && (row.global_model_name == requested_model_name
-            || (row_default_provider_model_name_available(row, api_format)
+            || (row_default_provider_model_name_available(row, api_format, request_operation)
                 && row.model_provider_model_name == requested_model_name)
             || row.global_model_mappings.as_ref().is_some_and(|patterns| {
                 patterns
@@ -105,7 +143,7 @@ fn row_supports_requested_model_exact(
             .as_ref()
             .is_some_and(|mappings| {
                 mappings.iter().any(|mapping| {
-                    mapping_scope_matches(mapping, row, api_format)
+                    mapping_scope_matches(mapping, row, api_format, request_operation)
                         && mapping.name == requested_model_name
                 })
             })
@@ -145,7 +183,24 @@ pub fn resolve_provider_model_name_with_model_directives(
     api_format: &str,
     enable_model_directives: bool,
 ) -> Option<(String, Option<String>)> {
-    let selected_provider_model_name = resolve_selected_provider_model_name(row, api_format)?;
+    resolve_provider_model_name_with_model_directives_and_request_operation(
+        row,
+        requested_model_name,
+        api_format,
+        enable_model_directives,
+        None,
+    )
+}
+
+pub fn resolve_provider_model_name_with_model_directives_and_request_operation(
+    row: &StoredMinimalCandidateSelectionRow,
+    requested_model_name: &str,
+    api_format: &str,
+    enable_model_directives: bool,
+    request_operation: Option<&str>,
+) -> Option<(String, Option<String>)> {
+    let selected_provider_model_name =
+        resolve_selected_provider_model_name(row, api_format, request_operation)?;
     let Some(key_allowed_models) = row.key_allowed_models.as_ref() else {
         return Some((selected_provider_model_name, None));
     };
@@ -175,7 +230,7 @@ pub fn resolve_provider_model_name_with_model_directives(
     sorted_allowed_models.sort_unstable();
 
     for &allowed_model in &sorted_allowed_models {
-        if row_has_candidate_model_name(row, api_format, allowed_model) {
+        if row_has_candidate_model_name(row, api_format, request_operation, allowed_model) {
             let allowed_model = allowed_model.to_owned();
             return Some((selected_provider_model_name.clone(), Some(allowed_model)));
         }
@@ -198,13 +253,14 @@ pub fn select_provider_model_name(
     row: &StoredMinimalCandidateSelectionRow,
     api_format: &str,
 ) -> String {
-    resolve_selected_provider_model_name(row, api_format)
+    resolve_selected_provider_model_name(row, api_format, None)
         .unwrap_or_else(|| row.model_provider_model_name.clone())
 }
 
 fn resolve_selected_provider_model_name(
     row: &StoredMinimalCandidateSelectionRow,
     api_format: &str,
+    request_operation: Option<&str>,
 ) -> Option<String> {
     let Some(mappings) = row.model_provider_model_mappings.as_ref() else {
         return Some(row.model_provider_model_name.clone());
@@ -212,17 +268,20 @@ fn resolve_selected_provider_model_name(
 
     if let Some(mapping) = mappings
         .iter()
-        .filter(|mapping| mapping_scope_matches(mapping, row, api_format))
+        .filter(|mapping| mapping_scope_matches(mapping, row, api_format, request_operation))
         .min_by(|left, right| {
             left.priority
                 .cmp(&right.priority)
+                .then_with(|| {
+                    mapping_operation_scope_rank(right).cmp(&mapping_operation_scope_rank(left))
+                })
                 .then(left.name.cmp(&right.name))
         })
     {
         return Some(mapping.name.clone());
     }
 
-    row_default_provider_model_name_available(row, api_format)
+    row_default_provider_model_name_available(row, api_format, request_operation)
         .then(|| row.model_provider_model_name.clone())
 }
 
@@ -231,12 +290,12 @@ pub fn candidate_model_names(
     api_format: &str,
 ) -> BTreeSet<String> {
     let mut names = BTreeSet::new();
-    if row_default_provider_model_name_available(row, api_format) {
+    if row_default_provider_model_name_available(row, api_format, None) {
         names.insert(row.model_provider_model_name.clone());
     }
     if let Some(mappings) = row.model_provider_model_mappings.as_ref() {
         for mapping in mappings {
-            if mapping_scope_matches(mapping, row, api_format) {
+            if mapping_scope_matches(mapping, row, api_format, None) {
                 names.insert(mapping.name.clone());
             }
         }
@@ -247,13 +306,15 @@ pub fn candidate_model_names(
 fn row_has_available_provider_model(
     row: &StoredMinimalCandidateSelectionRow,
     api_format: &str,
+    request_operation: Option<&str>,
 ) -> bool {
-    resolve_selected_provider_model_name(row, api_format).is_some()
+    resolve_selected_provider_model_name(row, api_format, request_operation).is_some()
 }
 
 fn row_default_provider_model_name_available(
     row: &StoredMinimalCandidateSelectionRow,
     api_format: &str,
+    request_operation: Option<&str>,
 ) -> bool {
     let Some(mappings) = row.model_provider_model_mappings.as_ref() else {
         return true;
@@ -264,7 +325,7 @@ fn row_default_provider_model_name_available(
             continue;
         }
         has_explicit_default_mapping = true;
-        if mapping_scope_matches(mapping, row, api_format) {
+        if mapping_scope_matches(mapping, row, api_format, request_operation) {
             return true;
         }
     }
@@ -275,6 +336,7 @@ fn mapping_scope_matches(
     mapping: &StoredProviderModelMapping,
     row: &StoredMinimalCandidateSelectionRow,
     api_format: &str,
+    request_operation: Option<&str>,
 ) -> bool {
     let api_format_matches_scope = mapping.api_formats.as_ref().is_none_or(|api_formats| {
         api_formats
@@ -285,13 +347,28 @@ fn mapping_scope_matches(
         return false;
     }
 
-    mapping.endpoint_ids.as_ref().is_none_or(|endpoint_ids| {
+    let endpoint_matches_scope = mapping.endpoint_ids.as_ref().is_none_or(|endpoint_ids| {
         endpoint_ids
             .iter()
             .any(|endpoint_id| endpoint_id == &row.endpoint_id)
+    });
+    if !endpoint_matches_scope {
+        return false;
+    }
+
+    mapping.operations.as_ref().is_none_or(|operations| {
+        request_operation.is_some_and(|request_operation| {
+            operations
+                .iter()
+                .any(|operation| operation.eq_ignore_ascii_case(request_operation))
+        })
     })
 }
 
+fn mapping_operation_scope_rank(mapping: &StoredProviderModelMapping) -> u8 {
+    u8::from(mapping.operations.is_some())
+}
+
 pub fn row_supports_required_capability(
     row: &StoredMinimalCandidateSelectionRow,
     required_capability: &str,
@@ -401,16 +478,18 @@ pub fn normalize_api_format(value: &str) -> String {
 fn row_has_candidate_model_name(
     row: &StoredMinimalCandidateSelectionRow,
     api_format: &str,
+    request_operation: Option<&str>,
     model_name: &str,
 ) -> bool {
-    (row_default_provider_model_name_available(row, api_format)
+    (row_default_provider_model_name_available(row, api_format, request_operation)
         && row.model_provider_model_name == model_name)
         || row
             .model_provider_model_mappings
             .as_ref()
             .is_some_and(|mappings| {
                 mappings.iter().any(|mapping| {
-                    mapping_scope_matches(mapping, row, api_format) && mapping.name == model_name
+                    mapping_scope_matches(mapping, row, api_format, request_operation)
+                        && mapping.name == model_name
                 })
             })
 }
@@ -484,6 +563,7 @@ mod tests {
     use super::{
         matches_model_mapping, resolve_provider_model_name,
         resolve_provider_model_name_with_model_directives,
+        resolve_provider_model_name_with_model_directives_and_request_operation,
         resolve_requested_global_model_name_with_model_directives, row_supports_requested_model,
         row_supports_requested_model_with_model_directives,
     };
@@ -518,6 +598,7 @@ mod tests {
             priority: 1,
             api_formats: Some(vec!["openai:chat".to_string()]),
             endpoint_ids: None,
+            operations: None,
         }]);
 
         let resolved = resolve_provider_model_name(&row, "gpt-5", "openai:chat")
@@ -581,6 +662,7 @@ mod tests {
             priority: 1,
             api_formats: Some(vec!["openai:responses".to_string()]),
             endpoint_ids: None,
+            operations: None,
         }]);
 
         assert!(row_supports_requested_model(
@@ -599,6 +681,7 @@ mod tests {
             priority: 1,
             api_formats: Some(vec!["openai:search".to_string()]),
             endpoint_ids: None,
+            operations: None,
         }]);
         assert!(!row_supports_requested_model(
             &row,
@@ -607,6 +690,51 @@ mod tests {
         ));
     }
 
+    #[test]
+    fn operation_scoped_mapping_overrides_generic_mapping_for_compaction() {
+        let mut row = sample_row("gpt-5.6-sol", "gpt-5.6-sol");
+        row.endpoint_api_format = "openai:responses".to_string();
+        row.model_provider_model_mappings = Some(vec![
+            StoredProviderModelMapping {
+                name: "gpt-5.6-sol".to_string(),
+                priority: 1,
+                api_formats: Some(vec!["openai:responses".to_string()]),
+                endpoint_ids: None,
+                operations: None,
+            },
+            StoredProviderModelMapping {
+                name: "gpt-5.6-terra".to_string(),
+                priority: 1,
+                api_formats: Some(vec!["openai:responses".to_string()]),
+                endpoint_ids: None,
+                operations: Some(vec!["compact".to_string()]),
+            },
+        ]);
+
+        assert_eq!(
+            resolve_provider_model_name_with_model_directives_and_request_operation(
+                &row,
+                "gpt-5.6-sol",
+                "openai:responses",
+                false,
+                None,
+            )
+            .map(|resolved| resolved.0),
+            Some("gpt-5.6-sol".to_string())
+        );
+        assert_eq!(
+            resolve_provider_model_name_with_model_directives_and_request_operation(
+                &row,
+                "gpt-5.6-sol",
+                "openai:responses",
+                false,
+                Some("compact"),
+            )
+            .map(|resolved| resolved.0),
+            Some("gpt-5.6-terra".to_string())
+        );
+    }
+
     #[test]
     fn model_directive_suffix_prefers_exact_model_before_base_fallback() {
         let exact = sample_row("gpt-5.4-high", "gpt-5.4-high-upstream");
@@ -685,6 +813,7 @@ mod tests {
             priority: 1,
             api_formats: None,
             endpoint_ids: Some(vec!["endpoint-openai".to_string()]),
+            operations: None,
         }]);
 
         assert!(!row_supports_requested_model(
diff --git a/crates/aether-usage-runtime/src/write.rs b/crates/aether-usage-runtime/src/write.rs
index aa6e572a5..f2ab5ca93 100644
--- a/crates/aether-usage-runtime/src/write.rs
+++ b/crates/aether-usage-runtime/src/write.rs
@@ -254,8 +254,13 @@ pub fn build_lifecycle_usage_seed(
     let model = context_string(context, "model")
         .or_else(|| non_empty_str(plan.model_name.as_deref()))
         .unwrap_or_else(|| "unknown".to_string());
-    let request_type =
-        infer_request_type_from_contracts(api_format.as_deref(), endpoint_api_format.as_deref());
+    let provider_request = context_body_value(context, "provider_request_body")
+        .or_else(|| plan_json_body_capture_for_usage(plan));
+    let request_type = infer_request_type_from_contracts(
+        api_format.as_deref(),
+        endpoint_api_format.as_deref(),
+        provider_request.as_ref(),
+    );
     let api_family = api_format
         .as_deref()
         .and_then(infer_api_family)
@@ -804,6 +809,7 @@ pub fn build_terminal_usage_context_seed(
     let request_type = infer_request_type_from_contracts(
         Some(client_contract.as_str()),
         Some(provider_contract.as_str()),
+        request_capture.provider_request.as_ref(),
     );
     let has_format_conversion = resolve_has_format_conversion(
         context,
@@ -1697,6 +1703,7 @@ fn build_usage_event_data_seed_with_detail(
     let request_type = Some(infer_request_type_from_contracts(
         api_format.as_deref(),
         endpoint_api_format.as_deref(),
+        request_capture.provider_request.as_ref(),
     ));
     let api_family = api_format
         .as_deref()
@@ -2467,7 +2474,20 @@ fn infer_request_type(api_format: Option<&str>) -> String {
 fn infer_request_type_from_contracts(
     client_api_format: Option<&str>,
     provider_api_format: Option<&str>,
+    provider_request: Option<&Value>,
 ) -> String {
+    let empty_body = Value::Null;
+    let provider_request = provider_request.unwrap_or(&empty_body);
+    for api_format in [provider_api_format, client_api_format]
+        .into_iter()
+        .flatten()
+    {
+        if let Some(operation) =
+            aether_ai_formats::openai_responses_request_operation(api_format, provider_request)
+        {
+            return operation.to_string();
+        }
+    }
     if matches!(
         infer_endpoint_kind(provider_api_format.unwrap_or_default()),
         Some("image")
@@ -3711,7 +3731,9 @@ mod tests {
                     "candidate_id": "cand-pending-event-1",
                     "candidate_index": 3,
                     "original_request_body": {"messages": [{"content": "omit me"}]},
-                    "provider_request_body": {"input": "omit me too"}
+                    "provider_request_body": {
+                        "input": [{"type": "compaction_trigger"}]
+                    }
                 })),
             ),
             1_700_000_020,
@@ -3723,6 +3745,7 @@ mod tests {
         assert_eq!(record.request_id, "req-pending-event-1");
         assert_eq!(record.status, "pending");
         assert_eq!(record.billing_status, "pending");
+        assert_eq!(record.request_type.as_deref(), Some("compact"));
         assert_eq!(record.finalized_at_unix_secs, None);
         assert_eq!(record.updated_at_unix_secs, 1_700_000_020);
         assert!(record.request_body.is_none());

From b09d1f1c336cb4902b1e9437d448b75fc6fbddc0 Mon Sep 17 00:00:00 2001
From: MMEXA <youlancelot@gmail.com>
Date: Sun, 12 Jul 2026 23:05:00 +0800
Subject: [PATCH 06/12] fix(usage): expose pending reasoning and exact reset
 expiry

---
 crates/aether-usage-runtime/src/write.rs      | 24 +++++++++++++++++--
 .../components/ProviderDetailDrawer.vue       |  6 ++---
 .../codex-reset-credit-display.spec.ts        | 20 ++++++++++++----
 .../components/codex-reset-credit-display.ts  | 12 ++++++----
 .../__tests__/UsageRecordsTable.spec.ts       |  9 +++++++
 5 files changed, 58 insertions(+), 13 deletions(-)

diff --git a/crates/aether-usage-runtime/src/write.rs b/crates/aether-usage-runtime/src/write.rs
index f2ab5ca93..751da9c00 100644
--- a/crates/aether-usage-runtime/src/write.rs
+++ b/crates/aether-usage-runtime/src/write.rs
@@ -2029,6 +2029,22 @@ fn build_runtime_request_metadata_seed(
         provider_request_body_ref.as_deref(),
         plan.body.body_bytes_b64.as_deref(),
     );
+    let provider_request_body = plan.body.json_body.as_ref().or_else(|| {
+        context_value_ref(context, "provider_request_body").filter(|value| !value.is_null())
+    });
+    let provider_api_format = context_string(context, "provider_api_format")
+        .or_else(|| non_empty_str(Some(plan.provider_api_format.as_str())));
+    let provider_model = context_string(context, "mapped_model")
+        .or_else(|| non_empty_str(plan.model_name.as_deref()));
+    let source_model =
+        context_string(context, "model").or_else(|| non_empty_str(plan.model_name.as_deref()));
+    metadata = attach_provider_request_body_metadata(
+        metadata,
+        provider_api_format.as_deref(),
+        provider_model.as_deref(),
+        source_model.as_deref(),
+        provider_request_body,
+    );
     if let Some(proxy) = plan.proxy.as_ref() {
         if let Some(node_id) = proxy
             .node_id
@@ -3625,7 +3641,8 @@ mod tests {
             content_encoding: None,
             body: RequestBody::from_json(json!({
                 "model": "gpt-5.4",
-                "messages": [{"role": "user", "content": "hello"}]
+                "messages": [{"role": "user", "content": "hello"}],
+                "reasoning": {"effort": "max"}
             })),
             stream: false,
             client_api_format: "claude:messages".to_string(),
@@ -3684,7 +3701,10 @@ mod tests {
             .as_ref()
             .and_then(Value::as_object)
             .expect("pending usage should only keep lightweight request metadata");
-        assert_eq!(metadata.len(), 1);
+        assert_eq!(
+            metadata.get("provider_reasoning_effort"),
+            Some(&json!("max"))
+        );
         let body_size = metadata
             .get("body_size")
             .and_then(Value::as_object)
diff --git a/frontend/src/features/providers/components/ProviderDetailDrawer.vue b/frontend/src/features/providers/components/ProviderDetailDrawer.vue
index aaa318121..d18c4f4cc 100644
--- a/frontend/src/features/providers/components/ProviderDetailDrawer.vue
+++ b/frontend/src/features/providers/components/ProviderDetailDrawer.vue
@@ -310,7 +310,7 @@
                                 :title="item.title"
                                 class="tabular-nums"
                               >
-                                {{ item.displayKey }} {{ formatCodexResetCreditDays(item.remainingSeconds) }}
+                                {{ item.displayKey }} {{ formatCodexResetCreditExpiresAt(item.expiresAt) }}
                               </span>
                               <span
                                 v-if="itemIndex < getVisibleCodexResetCreditItems(key).length - 1"
@@ -1019,7 +1019,7 @@ import {
 import { getGeminiCliAccountCreditsText } from '@/utils/providerKeyQuota'
 import {
   formatCodexResetCreditCount as formatCodexResetCreditCountLabel,
-  formatCodexResetCreditDays,
+  formatCodexResetCreditExpiresAt,
   getCodexResetCreditAvailableCount as getCodexResetCreditAvailableCountFromSnapshot,
   getVisibleCodexResetCreditItems as getVisibleCodexResetCreditItemsFromSnapshot,
 } from './codex-reset-credit-display'
@@ -1702,7 +1702,7 @@ async function handleConsumeCodexResetCredit(key: EndpointAPIKey) {
 
   const earliest = getVisibleCodexResetCreditItems(key)[0]
   const detailMessage = earliest
-    ? `\n当前最早过期项：${earliest.displayKey}，约 ${formatCodexResetCreditDays(earliest.remainingSeconds)} 后过期。`
+    ? `\n当前最早过期项：${earliest.displayKey}，${formatCodexResetCreditExpiresAt(earliest.expiresAt)} 过期。`
     : ''
   const confirmed = await confirm({
     title: legacyT('确认使用 Codex 重置机会'),
diff --git a/frontend/src/features/providers/components/__tests__/codex-reset-credit-display.spec.ts b/frontend/src/features/providers/components/__tests__/codex-reset-credit-display.spec.ts
index 82e877896..2fe140f70 100644
--- a/frontend/src/features/providers/components/__tests__/codex-reset-credit-display.spec.ts
+++ b/frontend/src/features/providers/components/__tests__/codex-reset-credit-display.spec.ts
@@ -2,7 +2,7 @@ import { describe, expect, it } from 'vitest'
 
 import {
   formatCodexResetCreditCount,
-  formatCodexResetCreditDays,
+  formatCodexResetCreditExpiresAt,
   getCodexResetCreditAvailableCount,
   getVisibleCodexResetCreditItems,
 } from '@/features/providers/components/codex-reset-credit-display'
@@ -74,8 +74,20 @@ describe('codex reset credit display helpers', () => {
     ])
   })
 
-  it('formats reset credit remaining days with a one-day minimum', () => {
-    expect(formatCodexResetCreditDays(1)).toBe('1天')
-    expect(formatCodexResetCreditDays(86_401)).toBe('2天')
+  it('formats reset credit expiry as a precise local timestamp', () => {
+    const expiresAt = new Date(2026, 6, 12, 22, 4, 41).getTime() / 1000
+    expect(formatCodexResetCreditExpiresAt(expiresAt)).toBe('07-12 22:04:41')
+    expect(formatCodexResetCreditExpiresAt(null)).toBe('-')
+  })
+
+  it('derives a stable expiry timestamp from remaining seconds', () => {
+    const snapshot: QuotaResetCreditsSnapshot = {
+      available_count: 1,
+      updated_at: 1_700_000_000,
+      credits: [{ status: 'available', remaining_seconds: 600 }],
+    }
+
+    expect(getVisibleCodexResetCreditItems(snapshot, 1_700_000_300)[0]?.expiresAt)
+      .toBe(1_700_000_600)
   })
 })
diff --git a/frontend/src/features/providers/components/codex-reset-credit-display.ts b/frontend/src/features/providers/components/codex-reset-credit-display.ts
index 9cdc9edc6..9b89c2bd9 100644
--- a/frontend/src/features/providers/components/codex-reset-credit-display.ts
+++ b/frontend/src/features/providers/components/codex-reset-credit-display.ts
@@ -66,7 +66,7 @@ export function getVisibleCodexResetCreditItems(
       if (remainingSeconds === null || remainingSeconds <= 0) return null
       return {
         id: item.id,
-        expiresAt: item.expires_at,
+        expiresAt: nowUnixSecs + remainingSeconds,
         remainingSeconds,
       } satisfies CodexResetCreditDisplayCandidate
     })
@@ -83,7 +83,11 @@ export function getVisibleCodexResetCreditItems(
     })
 }
 
-export function formatCodexResetCreditDays(remainingSeconds: number): string {
-  const days = Math.max(1, Math.ceil(remainingSeconds / 86_400))
-  return `${days}天`
+export function formatCodexResetCreditExpiresAt(expiresAt: number | null | undefined): string {
+  if (typeof expiresAt !== 'number' || !Number.isFinite(expiresAt)) return '-'
+  const date = new Date(expiresAt * 1000)
+  if (Number.isNaN(date.getTime())) return '-'
+
+  const pad = (value: number) => String(value).padStart(2, '0')
+  return `${pad(date.getMonth() + 1)}-${pad(date.getDate())} ${pad(date.getHours())}:${pad(date.getMinutes())}:${pad(date.getSeconds())}`
 }
diff --git a/frontend/src/features/usage/components/__tests__/UsageRecordsTable.spec.ts b/frontend/src/features/usage/components/__tests__/UsageRecordsTable.spec.ts
index e399faec7..79fa4bec1 100644
--- a/frontend/src/features/usage/components/__tests__/UsageRecordsTable.spec.ts
+++ b/frontend/src/features/usage/components/__tests__/UsageRecordsTable.spec.ts
@@ -288,6 +288,15 @@ describe('UsageRecordsTable', () => {
     expect(root.textContent).toContain('xhigh')
   })
 
+  it('shows request reasoning effort while the record is pending', () => {
+    const root = mountUsageRecordsTable([buildRecord({
+      status: 'pending',
+      reasoning_effort: 'max',
+    })])
+
+    expect(root.textContent).toContain('max')
+  })
+
   it('shows fast badge for priority service tier', () => {
     const root = mountUsageRecordsTable([buildRecord({ service_tier: 'priority' })])
 

From 3f86fdd6bc41333dcd872060e50345b8e5e805f4 Mon Sep 17 00:00:00 2001
From: MMEXA <youlancelot@gmail.com>
Date: Mon, 13 Jul 2026 21:45:53 +0800
Subject: [PATCH 07/12] =?UTF-8?q?feat(usage):=20=E5=B1=95=E7=A4=BA?=
 =?UTF-8?q?=E5=8E=8B=E7=BC=A9=E6=93=8D=E4=BD=9C=E4=B8=8E=E8=BF=9B=E8=A1=8C?=
 =?UTF-8?q?=E6=80=81=E8=AF=B7=E6=B1=82=E8=AF=AD=E4=B9=89?=
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

---
 frontend/src/api/endpoints/types/provider.ts  |  1 +
 frontend/src/api/me.ts                        |  2 +
 frontend/src/api/usage.ts                     |  2 +
 .../components/ModelMappingDialog.vue         | 76 ++++++++++++++++++-
 .../provider-tabs/ModelAliasesTab.vue         | 36 +++++++--
 .../provider-tabs/ModelMappingTab.vue         | 35 +++++++--
 .../usage/components/UsageRecordsTable.vue    | 31 +++++++-
 .../__tests__/UsageRecordsTable.spec.ts       |  9 +++
 .../__tests__/useUsageData.spec.ts            |  3 +
 .../usage/composables/useUsageData.ts         |  1 +
 frontend/src/features/usage/types.ts          |  1 +
 frontend/src/views/shared/Usage.vue           |  5 ++
 12 files changed, 188 insertions(+), 14 deletions(-)

diff --git a/frontend/src/api/endpoints/types/provider.ts b/frontend/src/api/endpoints/types/provider.ts
index d0580dcea..75c081704 100644
--- a/frontend/src/api/endpoints/types/provider.ts
+++ b/frontend/src/api/endpoints/types/provider.ts
@@ -940,6 +940,7 @@ export interface ProviderModelMapping {
   priority: number  // 优先级（数字越小优先级越高）
   api_formats?: string[]  // 作用域（适用的 API 格式），为空表示对所有格式生效
   endpoint_ids?: string[]  // 作用域（适用的端点 ID），为空表示对所有端点生效
+  operations?: string[]  // 作用域（适用的请求操作），为空表示对该格式的全部操作生效
 }
 
 // 保留别名以保持向后兼容
diff --git a/frontend/src/api/me.ts b/frontend/src/api/me.ts
index 39efc1a28..9ec11c9d2 100644
--- a/frontend/src/api/me.ts
+++ b/frontend/src/api/me.ts
@@ -54,6 +54,7 @@ export interface UsageRecordDetail {
   id: string
   provider?: string // 仅管理员可见
   model: string
+  request_type?: string | null
   reasoning_effort?: string | null
   service_tier?: string | null
   actual_service_tier?: string | null
@@ -368,6 +369,7 @@ export const meApi = {
       has_format_conversion?: boolean | null
       has_fallback?: boolean | null
       target_model?: string | null
+      request_type?: string | null
       reasoning_effort?: string | null
       service_tier?: string | null
       actual_service_tier?: string | null
diff --git a/frontend/src/api/usage.ts b/frontend/src/api/usage.ts
index 38a55f294..35385eb3c 100644
--- a/frontend/src/api/usage.ts
+++ b/frontend/src/api/usage.ts
@@ -14,6 +14,7 @@ export interface UsageRecord {
   provider_id?: string // UUID
   provider_name?: string
   model: string
+  request_type?: string | null
   reasoning_effort?: string | null
   service_tier?: string | null
   actual_service_tier?: string | null
@@ -566,6 +567,7 @@ export const usageApi = {
       has_format_conversion?: boolean | null
       has_fallback?: boolean | null
       target_model?: string | null
+      request_type?: string | null
       reasoning_effort?: string | null
       service_tier?: string | null
       actual_service_tier?: string | null
diff --git a/frontend/src/features/providers/components/ModelMappingDialog.vue b/frontend/src/features/providers/components/ModelMappingDialog.vue
index 1584e0925..80c32246f 100644
--- a/frontend/src/features/providers/components/ModelMappingDialog.vue
+++ b/frontend/src/features/providers/components/ModelMappingDialog.vue
@@ -58,6 +58,21 @@
         </p>
       </div>
 
+      <div class="space-y-1.5">
+        <div class="flex items-center justify-between gap-2">
+          <Label class="text-xs">请求操作</Label>
+          <span class="text-xs text-muted-foreground">{{ operationScopeSummary }}</span>
+        </div>
+        <MultiSelect
+          v-model="selectedOperations"
+          :options="operationOptions"
+          placeholder="全部操作"
+          empty-text="暂无可选操作"
+          no-results-text="未找到操作"
+          trigger-class="h-9 rounded-md"
+        />
+      </div>
+
       <!-- 映射名称选择面板 -->
       <div class="space-y-1.5">
         <Label class="text-xs">提供商模型</Label>
@@ -302,6 +317,8 @@ export interface AliasGroup {
   apiFormats: string[]
   endpointIdsKey: string
   endpointIds: string[]
+  operationsKey: string
+  operations: string[]
   aliases: ProviderModelAlias[]
 }
 
@@ -330,6 +347,11 @@ type EndpointOption = {
   label: string
 }
 
+type OperationOption = {
+  value: string
+  label: string
+}
+
 // 状态
 const submitting = ref(false)
 const loadingModels = ref(false)
@@ -358,6 +380,12 @@ const selectedNames = ref<string[]>([])
 // 选中的端点 ID；空数组表示全部端点
 const selectedEndpointIds = ref<string[]>([])
 
+const selectedOperations = ref<string[]>([])
+
+const operationOptions: OperationOption[] = [
+  { value: 'compact', label: '线程压缩' }
+]
+
 // 自定义名称列表（手动添加的）
 const allCustomNames = ref<string[]>([])
 
@@ -391,6 +419,19 @@ const endpointScopeSummary = computed(() => {
   return `${selected.length} 个端点`
 })
 
+const normalizedSelectedOperations = computed(() => {
+  const selected = normalizeStringList(selectedOperations.value)
+  return selected.length > 0 ? selected : undefined
+})
+
+const operationScopeSummary = computed(() => {
+  const selected = normalizedSelectedOperations.value
+  if (!selected) return '全部操作'
+  return selected.length === 1 && selected[0] === 'compact'
+    ? '线程压缩'
+    : `${selected.length} 项操作`
+})
+
 // 所有已知名称集合
 const allKnownNames = computed(() => {
   const set = new Set<string>()
@@ -523,6 +564,7 @@ function findDuplicateNames(
   names: string[],
   endpointIds: string[] | undefined,
   apiFormats: string[] | undefined = undefined,
+  operations: string[] | undefined = undefined,
 ): string[] {
   const duplicates = new Set<string>()
   for (const rawName of names) {
@@ -532,6 +574,7 @@ function findDuplicateNames(
       return alias.name === name
         && scopesOverlap(alias.endpoint_ids, endpointIds)
         && scopesOverlap(alias.api_formats, apiFormats)
+        && scopesOverlap(alias.operations, operations)
     })
     if (duplicate) duplicates.add(name)
   }
@@ -597,6 +640,7 @@ function initForm() {
     const existingNames = props.editingGroup.aliases.map(a => a.name)
     selectedNames.value = [...existingNames]
     selectedEndpointIds.value = normalizeStringList(props.editingGroup.endpointIds)
+    selectedOperations.value = normalizeStringList(props.editingGroup.operations)
     allCustomNames.value = [...existingNames]
   } else {
     formData.value = {
@@ -604,6 +648,7 @@ function initForm() {
     }
     selectedNames.value = []
     selectedEndpointIds.value = []
+    selectedOperations.value = []
     allCustomNames.value = []
   }
   searchQuery.value = ''
@@ -626,6 +671,10 @@ function getEndpointIdsKey(endpointIds: string[] | undefined): string {
   return getScopeKey(endpointIds)
 }
 
+function getOperationsKey(operations: string[] | undefined): string {
+  return getScopeKey(operations)
+}
+
 // 提交表单
 async function handleSubmit() {
   if (submitting.value) return
@@ -642,6 +691,7 @@ async function handleSubmit() {
     const currentAliases = targetModel.provider_model_mappings || []
     let newAliases: ProviderModelAlias[]
     const nextEndpointIds = normalizedSelectedEndpointIds.value
+    const nextOperations = normalizedSelectedOperations.value
 
     const buildAliases = (names: string[]): ProviderModelAlias[] => {
       return names.map((name) => {
@@ -652,6 +702,9 @@ async function handleSubmit() {
         if (nextEndpointIds && nextEndpointIds.length > 0) {
           alias.endpoint_ids = nextEndpointIds
         }
+        if (nextOperations && nextOperations.length > 0) {
+          alias.operations = nextOperations
+        }
         return alias
       })
     }
@@ -659,15 +712,26 @@ async function handleSubmit() {
     if (props.editingGroup) {
       const oldApiFormatsKey = props.editingGroup.apiFormatsKey
       const oldEndpointIdsKey = props.editingGroup.endpointIdsKey
+      const oldOperationsKey = props.editingGroup.operationsKey
       const oldAliasNames = new Set(props.editingGroup.aliases.map(a => a.name))
 
       const filteredAliases = currentAliases.filter((a: ProviderModelAlias) => {
         const currentKey = getApiFormatsKey(a.api_formats)
         const currentEndpointIdsKey = getEndpointIdsKey(a.endpoint_ids)
-        return !(currentKey === oldApiFormatsKey && currentEndpointIdsKey === oldEndpointIdsKey && oldAliasNames.has(a.name))
+        const currentOperationsKey = getOperationsKey(a.operations)
+        return !(currentKey === oldApiFormatsKey
+          && currentEndpointIdsKey === oldEndpointIdsKey
+          && currentOperationsKey === oldOperationsKey
+          && oldAliasNames.has(a.name))
       })
 
-      const duplicates = findDuplicateNames(filteredAliases, selectedNames.value, nextEndpointIds)
+      const duplicates = findDuplicateNames(
+        filteredAliases,
+        selectedNames.value,
+        nextEndpointIds,
+        undefined,
+        nextOperations,
+      )
       if (duplicates.length > 0) {
         showError(`以下映射名称已存在：${duplicates.join(', ')}`, '错误')
         return
@@ -678,7 +742,13 @@ async function handleSubmit() {
         ...buildAliases(selectedNames.value)
       ]
     } else {
-      const duplicates = findDuplicateNames(currentAliases, selectedNames.value, nextEndpointIds)
+      const duplicates = findDuplicateNames(
+        currentAliases,
+        selectedNames.value,
+        nextEndpointIds,
+        undefined,
+        nextOperations,
+      )
       if (duplicates.length > 0) {
         showError(`以下映射名称已存在：${duplicates.join(', ')}`, '错误')
         return
diff --git a/frontend/src/features/providers/components/provider-tabs/ModelAliasesTab.vue b/frontend/src/features/providers/components/provider-tabs/ModelAliasesTab.vue
index 839f22272..75dcfb907 100644
--- a/frontend/src/features/providers/components/provider-tabs/ModelAliasesTab.vue
+++ b/frontend/src/features/providers/components/provider-tabs/ModelAliasesTab.vue
@@ -75,6 +75,13 @@
               >
                 {{ getEndpointScopeLabel(group) }}
               </Badge>
+              <Badge
+                v-if="group.operations.length > 0"
+                variant="outline"
+                class="text-xs"
+              >
+                {{ getOperationScopeLabel(group) }}
+              </Badge>
             </div>
             <!-- 映射数量 -->
             <span class="text-xs text-muted-foreground shrink-0">
@@ -130,6 +137,7 @@
               </div>
               <!-- 测试按钮 -->
               <Button
+                v-if="group.operations.length === 0"
                 variant="ghost"
                 size="icon"
                 class="h-7 w-7 shrink-0"
@@ -269,8 +277,12 @@ function getEndpointIdsKey(endpointIds: string[] | undefined): string {
   return getScopeKey(endpointIds)
 }
 
+function getOperationsKey(operations: string[] | undefined): string {
+  return getScopeKey(operations)
+}
+
 function getAliasGroupKey(group: AliasGroup): string {
-  return `${group.model.id}-${group.apiFormatsKey}-${group.endpointIdsKey}`
+  return `${group.model.id}-${group.apiFormatsKey}-${group.endpointIdsKey}-${group.operationsKey}`
 }
 
 function getEndpointScopeLabel(group: AliasGroup): string {
@@ -278,6 +290,11 @@ function getEndpointScopeLabel(group: AliasGroup): string {
   return `${group.endpointIds.length} 端点`
 }
 
+function getOperationScopeLabel(group: AliasGroup): string {
+  if (group.operations.length === 1 && group.operations[0] === 'compact') return '压缩'
+  return `${group.operations.length} 项操作`
+}
+
 // 按"模型+作用域"分组的映射列表
 const aliasGroups = computed<AliasGroup[]>(() => {
   const groups: AliasGroup[] = []
@@ -289,7 +306,8 @@ const aliasGroups = computed<AliasGroup[]>(() => {
     for (const alias of model.provider_model_mappings) {
       const apiFormatsKey = getApiFormatsKey(alias.api_formats)
       const endpointIdsKey = getEndpointIdsKey(alias.endpoint_ids)
-      const groupKey = `${model.id}|${apiFormatsKey}|${endpointIdsKey}`
+      const operationsKey = getOperationsKey(alias.operations)
+      const groupKey = `${model.id}|${apiFormatsKey}|${endpointIdsKey}|${operationsKey}`
 
       if (!groupMap.has(groupKey)) {
         const group: AliasGroup = {
@@ -298,6 +316,8 @@ const aliasGroups = computed<AliasGroup[]>(() => {
           apiFormats: alias.api_formats || [],
           endpointIdsKey,
           endpointIds: normalizeStringList(alias.endpoint_ids),
+          operationsKey,
+          operations: normalizeStringList(alias.operations),
           aliases: []
         }
         groupMap.set(groupKey, group)
@@ -317,6 +337,7 @@ const aliasGroups = computed<AliasGroup[]>(() => {
     if (nameA !== nameB) return nameA.localeCompare(nameB)
     return a.apiFormatsKey.localeCompare(b.apiFormatsKey)
       || a.endpointIdsKey.localeCompare(b.endpointIdsKey)
+      || a.operationsKey.localeCompare(b.operationsKey)
   })
 })
 
@@ -339,8 +360,9 @@ const deleteConfirmDescription = computed(() => {
   const modelName = model.global_model_display_name || model.provider_model_name
   const scopeText = apiFormats.length === 0 ? '全部' : apiFormats.map(f => formatApiFormat(f)).join(', ')
   const endpointScope = getEndpointScopeLabel(deletingGroup.value)
+  const operationScope = getOperationScopeLabel(deletingGroup.value)
   const aliasNames = aliases.map(a => a.name).join(', ')
-  return `确定要删除模型「${modelName}」在作用域「${scopeText} / ${endpointScope}」下的 ${aliases.length} 个映射吗？\n\n映射名称：${aliasNames}`
+  return `确定要删除模型「${modelName}」在作用域「${scopeText} / ${endpointScope} / ${operationScope}」下的 ${aliases.length} 个映射吗？\n\n映射名称：${aliasNames}`
 })
 
 // 切换映射组展开状态
@@ -383,7 +405,7 @@ function deleteGroup(group: AliasGroup) {
 async function confirmDelete() {
   if (!deletingGroup.value) return
 
-  const { model, aliases, apiFormatsKey, endpointIdsKey } = deletingGroup.value
+  const { model, aliases, apiFormatsKey, endpointIdsKey, operationsKey } = deletingGroup.value
 
   try {
     const currentAliases = model.provider_model_mappings || []
@@ -391,7 +413,11 @@ async function confirmDelete() {
     const newAliases = currentAliases.filter((a: ProviderModelAlias) => {
       const currentKey = getApiFormatsKey(a.api_formats)
       const currentEndpointIdsKey = getEndpointIdsKey(a.endpoint_ids)
-      return !(currentKey === apiFormatsKey && currentEndpointIdsKey === endpointIdsKey && aliasNamesToRemove.has(a.name))
+      const currentOperationsKey = getOperationsKey(a.operations)
+      return !(currentKey === apiFormatsKey
+        && currentEndpointIdsKey === endpointIdsKey
+        && currentOperationsKey === operationsKey
+        && aliasNamesToRemove.has(a.name))
     })
 
     await updateModel(props.provider.id, model.id, {
diff --git a/frontend/src/features/providers/components/provider-tabs/ModelMappingTab.vue b/frontend/src/features/providers/components/provider-tabs/ModelMappingTab.vue
index 7a0b99337..d589d62cf 100644
--- a/frontend/src/features/providers/components/provider-tabs/ModelMappingTab.vue
+++ b/frontend/src/features/providers/components/provider-tabs/ModelMappingTab.vue
@@ -79,6 +79,13 @@
               >
                 {{ getGroupEndpointScopeLabel(item.group) }}
               </Badge>
+              <Badge
+                v-if="item.group && item.group.operations.length > 0"
+                variant="outline"
+                class="text-xs shrink-0"
+              >
+                {{ getGroupOperationScopeLabel(item.group) }}
+              </Badge>
             </template>
             <!-- 正则映射 -->
             <template v-else>
@@ -160,6 +167,7 @@
               </span>
               <!-- 测试按钮（直连测试） -->
               <Button
+                v-if="!item.group || item.group.operations.length === 0"
                 variant="ghost"
                 size="icon"
                 class="h-7 w-7 shrink-0"
@@ -492,11 +500,20 @@ function getEndpointIdsKey(endpointIds: string[] | undefined): string {
   return getScopeKey(endpointIds)
 }
 
+function getOperationsKey(operations: string[] | undefined): string {
+  return getScopeKey(operations)
+}
+
 function getGroupEndpointScopeLabel(group: AliasGroup): string {
   if (!group.endpointIds || group.endpointIds.length === 0) return '全部端点'
   return `${group.endpointIds.length} 端点`
 }
 
+function getGroupOperationScopeLabel(group: AliasGroup): string {
+  if (group.operations.length === 1 && group.operations[0] === 'compact') return '压缩'
+  return `${group.operations.length} 项操作`
+}
+
 // 精确映射分组（来自 provider_model_mappings）
 const exactMappingGroups = computed<AliasGroup[]>(() => {
   const groups: AliasGroup[] = []
@@ -508,7 +525,8 @@ const exactMappingGroups = computed<AliasGroup[]>(() => {
     for (const alias of model.provider_model_mappings) {
       const apiFormatsKey = getApiFormatsKey(alias.api_formats)
       const endpointIdsKey = getEndpointIdsKey(alias.endpoint_ids)
-      const groupKey = `${model.id}|${apiFormatsKey}|${endpointIdsKey}`
+      const operationsKey = getOperationsKey(alias.operations)
+      const groupKey = `${model.id}|${apiFormatsKey}|${endpointIdsKey}|${operationsKey}`
 
       if (!groupMap.has(groupKey)) {
         const group: AliasGroup = {
@@ -517,6 +535,8 @@ const exactMappingGroups = computed<AliasGroup[]>(() => {
           apiFormats: alias.api_formats || [],
           endpointIdsKey,
           endpointIds: normalizeStringList(alias.endpoint_ids),
+          operationsKey,
+          operations: normalizeStringList(alias.operations),
           aliases: []
         }
         groupMap.set(groupKey, group)
@@ -593,7 +613,7 @@ const combinedMappings = computed<CombinedMapping[]>(() => {
   // 添加精确映射
   for (const group of exactMappingGroups.value) {
     result.push({
-      key: `exact-${group.model.id}-${group.apiFormatsKey}`,
+      key: `exact-${group.model.id}-${group.apiFormatsKey}-${group.endpointIdsKey}-${group.operationsKey}`,
       type: 'exact',
       targetModelName: group.model.global_model_display_name || group.model.provider_model_name,
       targetModelId: group.model.id,
@@ -638,7 +658,8 @@ const deleteConfirmDescription = computed(() => {
   const modelName = model.global_model_display_name || model.provider_model_name
   const aliasNames = aliases.map(a => a.name).join(', ')
   const endpointScope = getGroupEndpointScopeLabel(deletingGroup.value)
-  return `确定要删除模型「${modelName}」在「${endpointScope}」下的 ${aliases.length} 个映射吗？\n\n映射名称：${aliasNames}`
+  const operationScope = getGroupOperationScopeLabel(deletingGroup.value)
+  return `确定要删除模型「${modelName}」在「${endpointScope} / ${operationScope}」下的 ${aliases.length} 个映射吗？\n\n映射名称：${aliasNames}`
 })
 
 // 切换展开状态
@@ -685,7 +706,7 @@ function deleteGroup(group: AliasGroup) {
 async function confirmDelete() {
   if (!deletingGroup.value) return
 
-  const { model, aliases, apiFormatsKey, endpointIdsKey } = deletingGroup.value
+  const { model, aliases, apiFormatsKey, endpointIdsKey, operationsKey } = deletingGroup.value
 
   try {
     const currentAliases = model.provider_model_mappings || []
@@ -693,7 +714,11 @@ async function confirmDelete() {
     const newAliases = currentAliases.filter((a: ProviderModelAlias) => {
       const currentKey = getApiFormatsKey(a.api_formats)
       const currentEndpointIdsKey = getEndpointIdsKey(a.endpoint_ids)
-      return !(currentKey === apiFormatsKey && currentEndpointIdsKey === endpointIdsKey && aliasNamesToRemove.has(a.name))
+      const currentOperationsKey = getOperationsKey(a.operations)
+      return !(currentKey === apiFormatsKey
+        && currentEndpointIdsKey === endpointIdsKey
+        && currentOperationsKey === operationsKey
+        && aliasNamesToRemove.has(a.name))
     })
 
     await updateModel(props.provider.id, model.id, {
diff --git a/frontend/src/features/usage/components/UsageRecordsTable.vue b/frontend/src/features/usage/components/UsageRecordsTable.vue
index 6ddfb03e0..83a0a8b37 100644
--- a/frontend/src/features/usage/components/UsageRecordsTable.vue
+++ b/frontend/src/features/usage/components/UsageRecordsTable.vue
@@ -244,6 +244,14 @@
           <div class="min-w-0 flex-1">
             <div class="flex min-w-0 items-center gap-1.5">
               <span class="min-w-0 truncate text-[15px] font-semibold leading-5">{{ record.model }}</span>
+              <Badge
+                v-if="getRequestTypeLabel(record)"
+                variant="outline"
+                class="h-4 rounded-full border-sky-500/30 bg-sky-500/5 px-1.5 text-[10px] leading-4 text-sky-700 dark:text-sky-300 flex-shrink-0"
+                title="线程压缩"
+              >
+                {{ getRequestTypeLabel(record) }}
+              </Badge>
               <Badge
                 v-if="getReasoningEffort(record)"
                 variant="outline"
@@ -753,6 +761,14 @@
                     clip-rule="evenodd"
                   />
                 </svg>
+                <Badge
+                  v-if="getRequestTypeLabel(record)"
+                  variant="outline"
+                  class="h-4 rounded-full border-sky-500/30 bg-sky-500/5 px-1.5 text-[10px] leading-4 text-sky-700 dark:text-sky-300 flex-shrink-0"
+                  title="线程压缩"
+                >
+                  {{ getRequestTypeLabel(record) }}
+                </Badge>
                 <Badge
                   v-if="getReasoningEffort(record)"
                   variant="outline"
@@ -777,6 +793,14 @@
               class="flex min-w-0 items-center gap-1"
             >
               <span class="truncate">{{ record.model }}</span>
+              <Badge
+                v-if="getRequestTypeLabel(record)"
+                variant="outline"
+                class="h-4 rounded-full border-sky-500/30 bg-sky-500/5 px-1.5 text-[10px] leading-4 text-sky-700 dark:text-sky-300 flex-shrink-0"
+                title="线程压缩"
+              >
+                {{ getRequestTypeLabel(record) }}
+              </Badge>
               <Badge
                 v-if="getReasoningEffort(record)"
                 variant="outline"
@@ -1605,6 +1629,10 @@ function getReasoningEffort(record: UsageRecord): string | null {
   return effort || null
 }
 
+function getRequestTypeLabel(record: UsageRecord): string | null {
+  return record.request_type?.trim().toLowerCase() === 'compact' ? '压缩' : null
+}
+
 function getReasoningEffortTitle(record: UsageRecord): string {
   const effort = getReasoningEffort(record)
   return effort ? `Reasoning: ${effort}` : ''
@@ -1644,8 +1672,9 @@ function getModelTooltip(record: UsageRecord): string {
   const actualModel = getActualModel(record)
   const reasoningEffort = getReasoningEffort(record)
   const serviceTierTitle = getFastBadgeTitle(record)
+  const requestType = getRequestTypeLabel(record)
   const tierSuffix = serviceTierTitle ? `\n${serviceTierTitle}` : ''
-  const suffix = `${reasoningEffort ? `\nReasoning: ${reasoningEffort}` : ''}${tierSuffix}`
+  const suffix = `${requestType ? `\n操作: ${requestType}` : ''}${reasoningEffort ? `\nReasoning: ${reasoningEffort}` : ''}${tierSuffix}`
   if (actualModel) {
     return `${record.model} -> ${actualModel}${suffix}`
   }
diff --git a/frontend/src/features/usage/components/__tests__/UsageRecordsTable.spec.ts b/frontend/src/features/usage/components/__tests__/UsageRecordsTable.spec.ts
index 79fa4bec1..27dd0dbee 100644
--- a/frontend/src/features/usage/components/__tests__/UsageRecordsTable.spec.ts
+++ b/frontend/src/features/usage/components/__tests__/UsageRecordsTable.spec.ts
@@ -297,6 +297,15 @@ describe('UsageRecordsTable', () => {
     expect(root.textContent).toContain('max')
   })
 
+  it('marks Responses compaction while the record is pending', () => {
+    const root = mountUsageRecordsTable([buildRecord({
+      status: 'pending',
+      request_type: 'compact',
+    })])
+
+    expect(root.textContent).toContain('压缩')
+  })
+
   it('shows fast badge for priority service tier', () => {
     const root = mountUsageRecordsTable([buildRecord({ service_tier: 'priority' })])
 
diff --git a/frontend/src/features/usage/composables/__tests__/useUsageData.spec.ts b/frontend/src/features/usage/composables/__tests__/useUsageData.spec.ts
index 987199ad2..d1e4dc2ba 100644
--- a/frontend/src/features/usage/composables/__tests__/useUsageData.spec.ts
+++ b/frontend/src/features/usage/composables/__tests__/useUsageData.spec.ts
@@ -216,6 +216,7 @@ describe('useUsageData', () => {
       has_format_conversion: false,
       has_retry: true,
       target_model: 'gpt-5.5',
+      request_type: 'compact',
       reasoning_effort: 'xhigh',
       service_tier: 'auto',
       actual_service_tier: 'priority',
@@ -245,6 +246,7 @@ describe('useUsageData', () => {
         has_format_conversion: undefined,
         has_retry: false,
         target_model: null,
+        request_type: null,
         reasoning_effort: null,
         service_tier: null,
         actual_service_tier: null,
@@ -279,6 +281,7 @@ describe('useUsageData', () => {
       has_format_conversion: false,
       has_retry: true,
       target_model: 'gpt-5.5',
+      request_type: 'compact',
       reasoning_effort: 'xhigh',
       service_tier: 'auto',
       actual_service_tier: 'priority',
diff --git a/frontend/src/features/usage/composables/useUsageData.ts b/frontend/src/features/usage/composables/useUsageData.ts
index 164327727..c1c8b4d4c 100644
--- a/frontend/src/features/usage/composables/useUsageData.ts
+++ b/frontend/src/features/usage/composables/useUsageData.ts
@@ -630,6 +630,7 @@ export function useUsageData(options: UseUsageDataOptions) {
         provider_key_name: record.provider_key_name || existing.provider_key_name,
         rate_multiplier: record.rate_multiplier ?? existing.rate_multiplier,
         target_model: record.target_model ?? existing.target_model,
+        request_type: record.request_type ?? existing.request_type,
         reasoning_effort: record.reasoning_effort ?? existing.reasoning_effort,
         service_tier: record.service_tier ?? existing.service_tier,
         actual_service_tier: record.actual_service_tier ?? existing.actual_service_tier
diff --git a/frontend/src/features/usage/types.ts b/frontend/src/features/usage/types.ts
index c2d9cfcbc..50f81fb1d 100644
--- a/frontend/src/features/usage/types.ts
+++ b/frontend/src/features/usage/types.ts
@@ -96,6 +96,7 @@ export interface UsageRecord {
   model: string
   target_model?: string | null  // 映射后的目标模型名（若无映射则为空）
   model_version?: string | null  // Provider 返回的实际模型版本（列表轻量字段）
+  request_type?: string | null  // 由请求语义识别出的操作类型
   reasoning_effort?: string | null  // 从发送给 Provider 的请求体提取的 reasoning 级别
   service_tier?: string | null  // 从发送给 Provider 的请求体提取的服务层级
   actual_service_tier?: string | null  // Provider 响应确认的实际服务层级
diff --git a/frontend/src/views/shared/Usage.vue b/frontend/src/views/shared/Usage.vue
index 7c1e9e364..30a35b64e 100644
--- a/frontend/src/views/shared/Usage.vue
+++ b/frontend/src/views/shared/Usage.vue
@@ -558,6 +558,11 @@ async function pollActiveRequests() {
         if ('target_model' in update && (typeof update.target_model === 'string' || update.target_model === null)) {
           record.target_model = update.target_model
         }
+        if ('request_type' in update) {
+          record.request_type = typeof update.request_type === 'string'
+            ? update.request_type
+            : null
+        }
         if ('reasoning_effort' in update) {
           record.reasoning_effort = typeof update.reasoning_effort === 'string'
             ? update.reasoning_effort

From cfc4894dab500c78c19111ed932cd001770f025a Mon Sep 17 00:00:00 2001
From: MMEXA <youlancelot@gmail.com>
Date: Tue, 14 Jul 2026 00:30:24 +0800
Subject: [PATCH 08/12] =?UTF-8?q?fix(usage):=20=E4=BF=9D=E7=95=99=E6=9C=80?=
 =?UTF-8?q?=E6=96=B0=E8=BF=9B=E8=A1=8C=E6=80=81=E7=94=9F=E5=91=BD=E5=91=A8?=
 =?UTF-8?q?=E6=9C=9F=E4=BA=8B=E4=BB=B6?=
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

---
 crates/aether-usage-runtime/src/runtime.rs | 89 +++++++++++++++++++++-
 1 file changed, 88 insertions(+), 1 deletion(-)

diff --git a/crates/aether-usage-runtime/src/runtime.rs b/crates/aether-usage-runtime/src/runtime.rs
index 38745b131..61d5ca725 100644
--- a/crates/aether-usage-runtime/src/runtime.rs
+++ b/crates/aether-usage-runtime/src/runtime.rs
@@ -326,7 +326,7 @@ impl LifecycleEventCoalescer {
             return false;
         };
         let should_emit = entry.generation == generation && !terminal_cancel_is_active(&entry, now);
-        if should_emit || !terminal_cancel_is_active(&entry, now) {
+        if should_emit {
             entries.remove(request_id);
         }
         Self::compact_locked(&mut entries, now);
@@ -2963,6 +2963,93 @@ mod tests {
         panic!("pending lifecycle usage event was not enqueued");
     }
 
+    #[tokio::test]
+    async fn stale_lifecycle_generation_keeps_the_current_generation_registered() {
+        let coalescer = super::LifecycleEventCoalescer::default();
+        let request_id = "req-lifecycle-generation";
+        let pending_generation = coalescer
+            .register(request_id.to_string())
+            .await
+            .expect("pending generation should register");
+        let streaming_generation = coalescer
+            .register(request_id.to_string())
+            .await
+            .expect("streaming generation should register");
+
+        assert!(!coalescer.should_emit(request_id, pending_generation).await);
+        assert!(
+            coalescer
+                .should_emit(request_id, streaming_generation)
+                .await
+        );
+    }
+
+    #[tokio::test]
+    async fn streaming_lifecycle_event_survives_a_superseded_pending_event() {
+        let config = UsageRuntimeConfig {
+            enabled: true,
+            queue_lifecycle_events: true,
+            lifecycle_enqueue_delay_ms: 40,
+            stream_key: "usage:events:test:pending-streaming-coalescing".to_string(),
+            consumer_group: "usage_consumers_test_pending_streaming_coalescing".to_string(),
+            consumer_block_ms: 1,
+            ..UsageRuntimeConfig::default()
+        };
+        let queue_runner: Arc<dyn RuntimeQueueStore> =
+            Arc::new(RuntimeState::memory(MemoryRuntimeStateConfig::default()));
+        let queue = UsageQueue::new(Arc::clone(&queue_runner), config.clone())
+            .expect("usage queue should build");
+        queue
+            .ensure_consumer_group()
+            .await
+            .expect("consumer group should initialize");
+        let store = CloneQueueConfiguredUsageStore {
+            records: Arc::new(Mutex::new(Vec::new())),
+            queue: queue_runner,
+        };
+        let runtime = UsageRuntime::new(config).expect("usage runtime should build");
+
+        runtime
+            .enqueue_or_write_lifecycle(
+                &store,
+                UsageEvent::new(
+                    UsageEventType::Pending,
+                    "req-pending-streaming-coalescing",
+                    UsageEventData {
+                        provider_name: "openai".to_string(),
+                        model: "gpt-5.6-sol".to_string(),
+                        ..UsageEventData::default()
+                    },
+                ),
+            )
+            .await;
+        sleep(Duration::from_millis(10)).await;
+        runtime
+            .enqueue_or_write_lifecycle(
+                &store,
+                UsageEvent::new(
+                    UsageEventType::Streaming,
+                    "req-pending-streaming-coalescing",
+                    UsageEventData {
+                        provider_name: "openai".to_string(),
+                        model: "gpt-5.6-sol".to_string(),
+                        ..UsageEventData::default()
+                    },
+                ),
+            )
+            .await;
+
+        sleep(Duration::from_millis(90)).await;
+        let entries = queue
+            .read_group("usage-test-consumer-pending-streaming-coalescing")
+            .await
+            .expect("queue read should succeed");
+        assert_eq!(entries.len(), 1);
+        let event = UsageEvent::from_stream_fields(&entries[0].fields)
+            .expect("queued usage event should parse");
+        assert_eq!(event.event_type, UsageEventType::Streaming);
+    }
+
     #[tokio::test]
     async fn pending_lifecycle_enqueue_can_be_delayed() {
         let config = UsageRuntimeConfig {

From f10d631a9c57e643d1cab85ae273b595a52e4b7f Mon Sep 17 00:00:00 2001
From: MMEXA <youlancelot@gmail.com>
Date: Tue, 14 Jul 2026 00:30:39 +0800
Subject: [PATCH 09/12] =?UTF-8?q?feat(frontend):=20=E6=BE=84=E6=B8=85?=
 =?UTF-8?q?=E6=A8=A1=E5=9E=8B=E6=98=A0=E5=B0=84=E9=80=82=E7=94=A8=E8=8C=83?=
 =?UTF-8?q?=E5=9B=B4?=
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

---
 .../types/__tests__/api-format.spec.ts        |   4 +-
 .../src/api/endpoints/types/api-format.ts     |   4 +-
 .../components/ModelMappingDialog.vue         | 222 ++++++++++++------
 .../__tests__/ModelMappingDialog.spec.ts      | 182 ++++++++++++++
 .../provider-tabs/ModelAliasesTab.vue         |  42 +++-
 .../provider-tabs/ModelMappingTab.vue         |  69 ++++--
 .../utils/__tests__/modelMappingScope.spec.ts |  74 ++++++
 .../providers/utils/modelMappingScope.ts      | 132 +++++++++++
 .../usage/components/UsageRecordsTable.vue    |   8 +-
 .../__tests__/UsageRecordsTable.spec.ts       |   2 +-
 frontend/src/i18n/messages.ts                 |  29 +++
 11 files changed, 657 insertions(+), 111 deletions(-)
 create mode 100644 frontend/src/features/providers/components/__tests__/ModelMappingDialog.spec.ts
 create mode 100644 frontend/src/features/providers/utils/__tests__/modelMappingScope.spec.ts
 create mode 100644 frontend/src/features/providers/utils/modelMappingScope.ts

diff --git a/frontend/src/api/endpoints/types/__tests__/api-format.spec.ts b/frontend/src/api/endpoints/types/__tests__/api-format.spec.ts
index 70b002c6c..74f6e6ade 100644
--- a/frontend/src/api/endpoints/types/__tests__/api-format.spec.ts
+++ b/frontend/src/api/endpoints/types/__tests__/api-format.spec.ts
@@ -50,9 +50,11 @@ describe('api format display helpers', () => {
     ])
   })
 
-  it('applies Responses to Search permissions in one direction', () => {
+  it('applies Responses companion permissions in one direction', () => {
     expect(apiFormatPermissionCovers('OPENAI_RESPONSES', 'openai:search')).toBe(true)
+    expect(apiFormatPermissionCovers('OPENAI_RESPONSES', 'openai:responses:compact')).toBe(true)
     expect(apiFormatPermissionCovers('openai:search', 'openai:responses')).toBe(false)
+    expect(apiFormatPermissionCovers('openai:responses:compact', 'openai:responses')).toBe(false)
   })
 
   it('formats embedding api format ids distinctly from chat formats', () => {
diff --git a/frontend/src/api/endpoints/types/api-format.ts b/frontend/src/api/endpoints/types/api-format.ts
index 636355b0c..177c1be79 100644
--- a/frontend/src/api/endpoints/types/api-format.ts
+++ b/frontend/src/api/endpoints/types/api-format.ts
@@ -235,7 +235,9 @@ export function apiFormatPermissionCovers(
   return Boolean(allowed)
     && Boolean(requested)
     && (allowed === requested
-      || (allowed === API_FORMATS.OPENAI_RESPONSES && requested === API_FORMATS.OPENAI_SEARCH))
+      || (allowed === API_FORMATS.OPENAI_RESPONSES
+        && (requested === API_FORMATS.OPENAI_RESPONSES_COMPACT
+          || requested === API_FORMATS.OPENAI_SEARCH)))
 }
 
 // 工具函数：按 family 分组并排序 API 格式数组
diff --git a/frontend/src/features/providers/components/ModelMappingDialog.vue b/frontend/src/features/providers/components/ModelMappingDialog.vue
index 80c32246f..533039941 100644
--- a/frontend/src/features/providers/components/ModelMappingDialog.vue
+++ b/frontend/src/features/providers/components/ModelMappingDialog.vue
@@ -37,42 +37,6 @@
         </p>
       </div>
 
-      <!-- 端点限制 -->
-      <div class="space-y-1.5">
-        <div class="flex items-center justify-between gap-2">
-          <Label class="text-xs">限制端点</Label>
-          <span class="text-xs text-muted-foreground">{{ endpointScopeSummary }}</span>
-        </div>
-        <MultiSelect
-          v-model="selectedEndpointIds"
-          :options="endpointOptions"
-          placeholder="全部端点"
-          empty-text="暂无端点"
-          no-results-text="未找到端点"
-          trigger-class="h-9 rounded-md"
-          dropdown-min-width="24rem"
-          :search-threshold="4"
-        />
-        <p class="text-xs text-muted-foreground">
-          默认对全部端点生效；选择端点后，此映射只在选中的端点上生效
-        </p>
-      </div>
-
-      <div class="space-y-1.5">
-        <div class="flex items-center justify-between gap-2">
-          <Label class="text-xs">请求操作</Label>
-          <span class="text-xs text-muted-foreground">{{ operationScopeSummary }}</span>
-        </div>
-        <MultiSelect
-          v-model="selectedOperations"
-          :options="operationOptions"
-          placeholder="全部操作"
-          empty-text="暂无可选操作"
-          no-results-text="未找到操作"
-          trigger-class="h-9 rounded-md"
-        />
-      </div>
-
       <!-- 映射名称选择面板 -->
       <div class="space-y-1.5">
         <Label class="text-xs">提供商模型</Label>
@@ -259,6 +223,67 @@
           </div>
         </div>
       </div>
+
+      <div class="space-y-3 border-t border-border/60 pt-4">
+        <div class="flex flex-col gap-1.5 sm:flex-row sm:items-start sm:justify-between sm:gap-4">
+          <div class="min-w-0 space-y-0.5">
+            <h3 class="text-sm font-medium text-foreground">
+              适用范围
+            </h3>
+            <p class="text-xs text-muted-foreground">
+              {{ t('providers.modelMapping.scope.matchHelp') }}
+            </p>
+          </div>
+          <span class="max-w-full break-words text-left text-xs text-muted-foreground sm:max-w-[55%] sm:text-right">
+            {{ mappingScopeSummary }}
+          </span>
+        </div>
+
+        <div class="grid gap-4 sm:grid-cols-2">
+          <div class="min-w-0 space-y-1.5">
+            <Label class="text-xs">适用端点</Label>
+            <MultiSelect
+              v-model="selectedEndpointIds"
+              :options="endpointOptions"
+              :placeholder="t('providers.modelMapping.scope.allEndpoints')"
+              empty-text="暂无端点"
+              no-results-text="未找到端点"
+              trigger-class="h-9 rounded-md"
+              :search-threshold="4"
+            />
+            <p class="text-xs leading-5 text-muted-foreground">
+              {{ t('providers.modelMapping.scope.endpointHelp') }}
+            </p>
+          </div>
+
+          <div class="min-w-0 space-y-1.5">
+            <Label class="text-xs">适用请求</Label>
+            <div
+              class="flex min-h-9 w-full flex-wrap gap-1 rounded-md bg-muted/50 p-1"
+              role="radiogroup"
+              aria-label="适用请求"
+            >
+              <Button
+                v-for="option in requestScopeOptions"
+                :key="option.value"
+                type="button"
+                size="sm"
+                :variant="requestScopeValue === option.value ? 'secondary' : 'ghost'"
+                class="h-7 min-w-0 flex-1 basis-[9rem] px-2.5"
+                role="radio"
+                :aria-checked="requestScopeValue === option.value"
+                :title="option.label"
+                @click="handleRequestScopeChange(option.value)"
+              >
+                <span class="truncate">{{ option.label }}</span>
+              </Button>
+            </div>
+            <p class="text-xs leading-5 text-muted-foreground">
+              {{ requestScopeDescription }}
+            </p>
+          </div>
+        </div>
+      </div>
     </div>
 
     <template #footer>
@@ -276,7 +301,7 @@
           v-if="submitting"
           class="w-4 h-4 mr-2 animate-spin"
         />
-        {{ editingGroup ? '保存' : '添加' }}
+        {{ editingGroup ? '保存映射' : '添加映射' }}
       </Button>
     </template>
   </Dialog>
@@ -298,6 +323,7 @@ import {
 } from '@/components/ui'
 import MultiSelect from '@/components/common/MultiSelect.vue'
 import { useToast } from '@/composables/useToast'
+import { useI18n } from '@/i18n'
 import { parseApiError } from '@/utils/errorParser'
 import {
   type Model,
@@ -306,8 +332,18 @@ import {
   type UpstreamModel,
 } from '@/api/endpoints'
 import { updateModel } from '@/api/endpoints/models'
-import { formatApiFormat } from '@/api/endpoints/types/api-format'
 import { useUpstreamModelsCache } from '../composables/useUpstreamModelsCache'
+import {
+  ALL_REQUESTS_SCOPE_VALUE,
+  COMPACT_REQUEST_SCOPE_VALUE,
+  formatModelMappingEndpointLabel,
+  formatModelMappingRequestScope,
+  modelMappingOperationsKey,
+  modelMappingOperationsFromScopeValue,
+  modelMappingRequestScopeOptions,
+  modelMappingRequestScopeValue,
+  normalizeModelMappingOperations,
+} from '../utils/modelMappingScope'
 
 export interface AliasGroup {
   model: Model
@@ -340,6 +376,7 @@ const emit = defineEmits<{
 }>()
 
 const { error: showError, success: showSuccess, warning: showWarning } = useToast()
+const { t } = useI18n()
 const { fetchModels: fetchCachedModels } = useUpstreamModelsCache()
 
 type EndpointOption = {
@@ -347,11 +384,6 @@ type EndpointOption = {
   label: string
 }
 
-type OperationOption = {
-  value: string
-  label: string
-}
-
 // 状态
 const submitting = ref(false)
 const loadingModels = ref(false)
@@ -382,54 +414,77 @@ const selectedEndpointIds = ref<string[]>([])
 
 const selectedOperations = ref<string[]>([])
 
-const operationOptions: OperationOption[] = [
-  { value: 'compact', label: '线程压缩' }
-]
-
 // 自定义名称列表（手动添加的）
 const allCustomNames = ref<string[]>([])
 
 const endpointOptions = computed<EndpointOption[]>(() => {
-  return (props.endpoints ?? []).map((endpoint) => {
-    const status = endpoint.is_active ? '' : '（停用）'
-    return {
-      value: endpoint.id,
-      label: `${formatApiFormat(endpoint.api_format)}${status}`,
-    }
-  })
+  const endpoints = props.endpoints ?? []
+  return endpoints.map(endpoint => ({
+    value: endpoint.id,
+    label: formatModelMappingEndpointLabel(endpoint, endpoints),
+  }))
 })
 
 const normalizedSelectedEndpointIds = computed(() => {
-  const validIds = new Set(endpointOptions.value.map(option => option.value))
   const selected = normalizeStringList(selectedEndpointIds.value)
-  if (selected.length === 0) {
-    return undefined
-  }
-  const invalidSelected = selected.filter(endpointId => !validIds.has(endpointId))
-  const selectedValidCount = selected.filter(endpointId => validIds.has(endpointId)).length
-  if (validIds.size > 0 && invalidSelected.length === 0 && selectedValidCount === validIds.size) {
-    return undefined
-  }
-  return selected
+  return selected.length > 0 ? selected : undefined
 })
 
 const endpointScopeSummary = computed(() => {
   const selected = normalizedSelectedEndpointIds.value
-  if (!selected || selected.length === 0) return '全部端点'
-  return `${selected.length} 个端点`
+  if (!selected || selected.length === 0) {
+    return t('providers.modelMapping.scope.allEndpoints')
+  }
+  if (selected.length === 1) {
+    return endpointOptions.value.find(option => option.value === selected[0])?.label
+      ?? t('providers.modelMapping.scope.endpointCount', { count: 1 })
+  }
+  return t('providers.modelMapping.scope.endpointCount', { count: selected.length })
 })
 
+const requestScopeLabels = computed(() => ({
+  allRequests: t('providers.modelMapping.scope.allRequests'),
+  sessionCompactionOnly: t('providers.modelMapping.scope.sessionCompactionOnly'),
+  customOperations: (operations: string[]) => t(
+    'providers.modelMapping.scope.customOperations',
+    { operations: operations.join(', ') },
+  ),
+}))
+
 const normalizedSelectedOperations = computed(() => {
-  const selected = normalizeStringList(selectedOperations.value)
+  const selected = normalizeModelMappingOperations(selectedOperations.value)
   return selected.length > 0 ? selected : undefined
 })
 
 const operationScopeSummary = computed(() => {
-  const selected = normalizedSelectedOperations.value
-  if (!selected) return '全部操作'
-  return selected.length === 1 && selected[0] === 'compact'
-    ? '线程压缩'
-    : `${selected.length} 项操作`
+  return formatModelMappingRequestScope(
+    normalizedSelectedOperations.value,
+    requestScopeLabels.value,
+  )
+})
+
+const mappingScopeSummary = computed(() => {
+  return `${endpointScopeSummary.value} · ${operationScopeSummary.value}`
+})
+
+const requestScopeValue = computed(() => {
+  return modelMappingRequestScopeValue(selectedOperations.value)
+})
+
+const requestScopeOptions = computed(() => {
+  return modelMappingRequestScopeOptions(selectedOperations.value, requestScopeLabels.value)
+})
+
+const requestScopeDescription = computed(() => {
+  if (requestScopeValue.value === ALL_REQUESTS_SCOPE_VALUE) {
+    return t('providers.modelMapping.scope.allRequestsDescription')
+  }
+  if (requestScopeValue.value === COMPACT_REQUEST_SCOPE_VALUE) {
+    return t('providers.modelMapping.scope.sessionCompactionDescription')
+  }
+  return t('providers.modelMapping.scope.customOperationsDescription', {
+    operations: normalizeModelMappingOperations(selectedOperations.value).join(', '),
+  })
 })
 
 // 所有已知名称集合
@@ -559,6 +614,17 @@ function scopesOverlap(left: string[] | undefined, right: string[] | undefined):
   return leftValues.some(value => rightSet.has(value))
 }
 
+function operationScopesOverlap(
+  left: string[] | undefined,
+  right: string[] | undefined,
+): boolean {
+  const leftValues = normalizeModelMappingOperations(left)
+  const rightValues = normalizeModelMappingOperations(right)
+  if (leftValues.length === 0 || rightValues.length === 0) return true
+  const rightSet = new Set(rightValues)
+  return leftValues.some(value => rightSet.has(value))
+}
+
 function findDuplicateNames(
   existingAliases: ProviderModelAlias[],
   names: string[],
@@ -574,7 +640,7 @@ function findDuplicateNames(
       return alias.name === name
         && scopesOverlap(alias.endpoint_ids, endpointIds)
         && scopesOverlap(alias.api_formats, apiFormats)
-        && scopesOverlap(alias.operations, operations)
+        && operationScopesOverlap(alias.operations, operations)
     })
     if (duplicate) duplicates.add(name)
   }
@@ -640,7 +706,7 @@ function initForm() {
     const existingNames = props.editingGroup.aliases.map(a => a.name)
     selectedNames.value = [...existingNames]
     selectedEndpointIds.value = normalizeStringList(props.editingGroup.endpointIds)
-    selectedOperations.value = normalizeStringList(props.editingGroup.operations)
+    selectedOperations.value = normalizeModelMappingOperations(props.editingGroup.operations)
     allCustomNames.value = [...existingNames]
   } else {
     formData.value = {
@@ -662,6 +728,10 @@ function handleModelChange(value: string) {
   formData.value.modelId = value
 }
 
+function handleRequestScopeChange(value: string) {
+  selectedOperations.value = modelMappingOperationsFromScopeValue(value) ?? []
+}
+
 // 生成作用域唯一键
 function getApiFormatsKey(formats: string[] | undefined): string {
   return getScopeKey(formats)
@@ -672,7 +742,7 @@ function getEndpointIdsKey(endpointIds: string[] | undefined): string {
 }
 
 function getOperationsKey(operations: string[] | undefined): string {
-  return getScopeKey(operations)
+  return modelMappingOperationsKey(operations)
 }
 
 // 提交表单
@@ -712,7 +782,7 @@ async function handleSubmit() {
     if (props.editingGroup) {
       const oldApiFormatsKey = props.editingGroup.apiFormatsKey
       const oldEndpointIdsKey = props.editingGroup.endpointIdsKey
-      const oldOperationsKey = props.editingGroup.operationsKey
+      const oldOperationsKey = modelMappingOperationsKey(props.editingGroup.operations)
       const oldAliasNames = new Set(props.editingGroup.aliases.map(a => a.name))
 
       const filteredAliases = currentAliases.filter((a: ProviderModelAlias) => {
diff --git a/frontend/src/features/providers/components/__tests__/ModelMappingDialog.spec.ts b/frontend/src/features/providers/components/__tests__/ModelMappingDialog.spec.ts
new file mode 100644
index 000000000..4885a125c
--- /dev/null
+++ b/frontend/src/features/providers/components/__tests__/ModelMappingDialog.spec.ts
@@ -0,0 +1,182 @@
+import { afterEach, describe, expect, it, vi } from 'vitest'
+import { createApp, defineComponent, h, nextTick, ref, type App } from 'vue'
+
+import ModelMappingDialog, { type AliasGroup } from '../ModelMappingDialog.vue'
+import type { Model, ProviderEndpoint } from '@/api/endpoints'
+import { updateModel } from '@/api/endpoints/models'
+
+vi.mock('@/components/ui', async () => {
+  const { defineComponent, h } = await import('vue')
+
+  const passthrough = (name: string, tag = 'div') => defineComponent({
+    name,
+    setup(_, { slots }) {
+      return () => h(tag, [slots.default?.(), slots.footer?.()])
+    },
+  })
+
+  return {
+    Button: defineComponent({
+      name: 'ButtonStub',
+      setup(_, { attrs, slots }) {
+        return () => h('button', { ...attrs, type: 'button' }, slots.default?.())
+      },
+    }),
+    Dialog: passthrough('DialogStub'),
+    Input: defineComponent({
+      name: 'InputStub',
+      props: { modelValue: String },
+      emits: ['update:modelValue'],
+      setup(props, { attrs, emit }) {
+        return () => h('input', {
+          ...attrs,
+          value: props.modelValue ?? '',
+          onInput: (event: Event) => emit(
+            'update:modelValue',
+            (event.target as HTMLInputElement).value,
+          ),
+        })
+      },
+    }),
+    Label: passthrough('LabelStub', 'label'),
+    Select: passthrough('SelectStub'),
+    SelectContent: passthrough('SelectContentStub'),
+    SelectItem: passthrough('SelectItemStub'),
+    SelectTrigger: passthrough('SelectTriggerStub'),
+    SelectValue: passthrough('SelectValueStub', 'span'),
+  }
+})
+
+vi.mock('@/components/common/MultiSelect.vue', async () => {
+  const { defineComponent, h } = await import('vue')
+  return {
+    default: defineComponent({
+      name: 'MultiSelectStub',
+      setup() {
+        return () => h('div')
+      },
+    }),
+  }
+})
+
+vi.mock('lucide-vue-next', async () => {
+  const { defineComponent, h } = await import('vue')
+  const Icon = defineComponent({
+    name: 'IconStub',
+    setup() {
+      return () => h('span')
+    },
+  })
+  return {
+    Check: Icon,
+    ChevronDown: Icon,
+    Loader2: Icon,
+    Plus: Icon,
+    RefreshCw: Icon,
+    Search: Icon,
+    Tag: Icon,
+    Zap: Icon,
+  }
+})
+
+vi.mock('@/api/endpoints/models', () => ({
+  updateModel: vi.fn().mockResolvedValue(undefined),
+}))
+
+vi.mock('@/composables/useToast', () => ({
+  useToast: () => ({
+    error: vi.fn(),
+    success: vi.fn(),
+    warning: vi.fn(),
+  }),
+}))
+
+vi.mock('../../composables/useUpstreamModelsCache', () => ({
+  useUpstreamModelsCache: () => ({
+    fetchModels: vi.fn(),
+  }),
+}))
+
+const mountedApps: Array<{ app: App, root: HTMLElement }> = []
+
+afterEach(() => {
+  vi.mocked(updateModel).mockClear()
+  for (const { app, root } of mountedApps.splice(0)) {
+    app.unmount()
+    root.remove()
+  }
+})
+
+describe('ModelMappingDialog', () => {
+  it('normalizes and replaces an edited compact operation scope', async () => {
+    const endpoint = {
+      id: 'endpoint-responses',
+      api_format: 'openai:responses',
+      base_url: 'https://api.example.com/v1',
+      is_active: true,
+    } as ProviderEndpoint
+    const model = {
+      id: 'model-sol',
+      provider_model_name: 'gpt-5.6-sol',
+      global_model_display_name: 'GPT-5.6 Sol',
+      provider_model_mappings: [{
+        name: 'gpt-5.6-luna',
+        priority: 1,
+        endpoint_ids: [endpoint.id],
+        operations: ['Compact'],
+      }],
+    } as Model
+    const editingGroup: AliasGroup = {
+      model,
+      apiFormatsKey: '',
+      apiFormats: [],
+      endpointIdsKey: endpoint.id,
+      endpointIds: [endpoint.id],
+      operationsKey: 'Compact',
+      operations: ['Compact'],
+      aliases: model.provider_model_mappings ?? [],
+    }
+    const open = ref(false)
+    const root = document.createElement('div')
+    document.body.appendChild(root)
+    const app = createApp(defineComponent({
+      setup() {
+        return () => h(ModelMappingDialog, {
+          open: open.value,
+          providerId: 'provider-1',
+          endpoints: [endpoint],
+          models: [model],
+          editingGroup,
+          'onUpdate:open': (value: boolean) => { open.value = value },
+        })
+      },
+    }))
+    app.mount(root)
+    mountedApps.push({ app, root })
+
+    open.value = true
+    await nextTick()
+    await nextTick()
+    expect(root.textContent).toContain('仅会话压缩')
+
+    const scopeButtons = [...root.querySelectorAll('button')]
+    scopeButtons.find(button => button.textContent?.includes('所有请求'))?.click()
+    await nextTick()
+    scopeButtons.find(button => button.textContent?.includes('仅会话压缩'))?.click()
+    await nextTick()
+    const saveButton = [...root.querySelectorAll('button')]
+      .find(button => button.textContent?.includes('保存映射'))
+    expect(saveButton).toBeDefined()
+    saveButton?.click()
+    await vi.waitFor(() => expect(updateModel).toHaveBeenCalledTimes(1))
+
+    expect(updateModel).toHaveBeenCalledWith('provider-1', 'model-sol', {
+      provider_model_mappings: [{
+        name: 'gpt-5.6-luna',
+        priority: 1,
+        endpoint_ids: [endpoint.id],
+        operations: ['compact'],
+      }],
+    })
+  })
+})
diff --git a/frontend/src/features/providers/components/provider-tabs/ModelAliasesTab.vue b/frontend/src/features/providers/components/provider-tabs/ModelAliasesTab.vue
index 75dcfb907..2e350b53d 100644
--- a/frontend/src/features/providers/components/provider-tabs/ModelAliasesTab.vue
+++ b/frontend/src/features/providers/components/provider-tabs/ModelAliasesTab.vue
@@ -38,21 +38,21 @@
       >
         <!-- 分组头部（可点击展开） -->
         <div
-          class="flex items-center justify-between px-4 py-3 hover:bg-muted/20 cursor-pointer"
+          class="flex items-start justify-between px-4 py-3 hover:bg-muted/20 cursor-pointer"
           @click="toggleAliasGroupExpand(getAliasGroupKey(group))"
         >
-          <div class="flex items-center gap-2 flex-1 min-w-0">
+          <div class="flex min-w-0 flex-1 flex-wrap items-center gap-x-2 gap-y-1.5">
             <!-- 展开/收起图标 -->
             <ChevronRight
               class="w-4 h-4 text-muted-foreground shrink-0 transition-transform"
               :class="{ 'rotate-90': expandedAliasGroups.has(getAliasGroupKey(group)) }"
             />
             <!-- 模型名称 -->
-            <span class="font-semibold text-sm truncate">
+            <span class="min-w-0 flex-[1_1_12rem] truncate text-sm font-semibold">
               {{ group.model.global_model_display_name || group.model.provider_model_name }}
             </span>
             <!-- 作用域标签 -->
-            <div class="flex items-center gap-1 shrink-0">
+            <div class="flex min-w-0 max-w-full flex-wrap items-center gap-1">
               <Badge
                 v-if="group.apiFormats.length === 0"
                 variant="outline"
@@ -76,11 +76,10 @@
                 {{ getEndpointScopeLabel(group) }}
               </Badge>
               <Badge
-                v-if="group.operations.length > 0"
                 variant="outline"
-                class="text-xs"
+                class="min-w-0 max-w-full text-xs"
               >
-                {{ getOperationScopeLabel(group) }}
+                <span class="truncate">{{ getOperationScopeLabel(group) }}</span>
               </Badge>
             </div>
             <!-- 映射数量 -->
@@ -214,10 +213,16 @@ import {
   type ProviderModelAlias
 } from '@/api/endpoints'
 import { updateModel } from '@/api/endpoints/models'
+import { useI18n } from '@/i18n'
 import { parseApiError, parseTestModelError } from '@/utils/errorParser'
 import { formatApiFormat } from '@/api/endpoints/types/api-format'
 import { buildExactModelMappingTestRequest } from './model-test-request'
 import type { ProviderWithEndpointsSummary } from '@/api/endpoints'
+import {
+  formatModelMappingRequestScope,
+  modelMappingOperationsKey,
+  normalizeModelMappingOperations,
+} from '../../utils/modelMappingScope'
 
 const props = defineProps<{
   provider: ProviderWithEndpointsSummary
@@ -228,6 +233,7 @@ const emit = defineEmits<{
 }>()
 
 const { error: showError, success: showSuccess } = useToast()
+const { t } = useI18n()
 
 // 状态
 const loading = ref(false)
@@ -278,21 +284,31 @@ function getEndpointIdsKey(endpointIds: string[] | undefined): string {
 }
 
 function getOperationsKey(operations: string[] | undefined): string {
-  return getScopeKey(operations)
+  return modelMappingOperationsKey(operations)
 }
 
+const requestScopeLabels = computed(() => ({
+  allRequests: t('providers.modelMapping.scope.allRequests'),
+  sessionCompactionOnly: t('providers.modelMapping.scope.sessionCompactionOnly'),
+  customOperations: (operations: string[]) => t(
+    'providers.modelMapping.scope.customOperations',
+    { operations: operations.join(', ') },
+  ),
+}))
+
 function getAliasGroupKey(group: AliasGroup): string {
   return `${group.model.id}-${group.apiFormatsKey}-${group.endpointIdsKey}-${group.operationsKey}`
 }
 
 function getEndpointScopeLabel(group: AliasGroup): string {
-  if (!group.endpointIds || group.endpointIds.length === 0) return '全部端点'
-  return `${group.endpointIds.length} 端点`
+  if (!group.endpointIds || group.endpointIds.length === 0) {
+    return t('providers.modelMapping.scope.allEndpoints')
+  }
+  return t('providers.modelMapping.scope.endpointCount', { count: group.endpointIds.length })
 }
 
 function getOperationScopeLabel(group: AliasGroup): string {
-  if (group.operations.length === 1 && group.operations[0] === 'compact') return '压缩'
-  return `${group.operations.length} 项操作`
+  return formatModelMappingRequestScope(group.operations, requestScopeLabels.value)
 }
 
 // 按"模型+作用域"分组的映射列表
@@ -317,7 +333,7 @@ const aliasGroups = computed<AliasGroup[]>(() => {
           endpointIdsKey,
           endpointIds: normalizeStringList(alias.endpoint_ids),
           operationsKey,
-          operations: normalizeStringList(alias.operations),
+          operations: normalizeModelMappingOperations(alias.operations),
           aliases: []
         }
         groupMap.set(groupKey, group)
diff --git a/frontend/src/features/providers/components/provider-tabs/ModelMappingTab.vue b/frontend/src/features/providers/components/provider-tabs/ModelMappingTab.vue
index d589d62cf..3371221ff 100644
--- a/frontend/src/features/providers/components/provider-tabs/ModelMappingTab.vue
+++ b/frontend/src/features/providers/components/provider-tabs/ModelMappingTab.vue
@@ -39,10 +39,10 @@
       >
         <!-- 行头部（可点击展开） -->
         <div
-          class="flex items-center justify-between px-4 py-3 hover:bg-muted/20 cursor-pointer"
+          class="flex items-start justify-between px-4 py-3 hover:bg-muted/20 cursor-pointer"
           @click="toggleExpand(item.key)"
         >
-          <div class="flex items-center gap-2 flex-1 min-w-0">
+          <div class="flex min-w-0 flex-1 flex-wrap items-center gap-x-2 gap-y-1.5">
             <!-- 展开/收起图标 -->
             <ChevronRight
               class="w-4 h-4 text-muted-foreground shrink-0 transition-transform self-start mt-0.5"
@@ -50,7 +50,7 @@
             />
             <!-- 精确映射 -->
             <template v-if="item.type === 'exact'">
-              <div class="flex flex-col min-w-0">
+              <div class="flex min-w-0 flex-[1_1_12rem] flex-col">
                 <span class="font-semibold text-sm truncate">
                   {{ item.targetModelName }}
                 </span>
@@ -75,21 +75,23 @@
               <Badge
                 v-if="item.group"
                 variant="outline"
-                class="text-xs shrink-0"
+                class="min-w-0 max-w-full text-xs"
+                :title="getGroupEndpointScopeTitle(item.group)"
               >
-                {{ getGroupEndpointScopeLabel(item.group) }}
+                <span class="truncate">{{ getGroupEndpointScopeLabel(item.group) }}</span>
               </Badge>
               <Badge
-                v-if="item.group && item.group.operations.length > 0"
+                v-if="item.group"
                 variant="outline"
-                class="text-xs shrink-0"
+                class="min-w-0 max-w-full text-xs"
+                :title="getGroupOperationScopeLabel(item.group)"
               >
-                {{ getGroupOperationScopeLabel(item.group) }}
+                <span class="truncate">{{ getGroupOperationScopeLabel(item.group) }}</span>
               </Badge>
             </template>
             <!-- 正则映射 -->
             <template v-else>
-              <div class="flex flex-col min-w-0">
+              <div class="flex min-w-0 flex-[1_1_12rem] flex-col">
                 <span class="font-semibold text-sm truncate">
                   {{ item.targetModelName }}
                 </span>
@@ -372,8 +374,15 @@ import {
 } from '@/api/endpoints'
 import { type EndpointAPIKey } from '@/api/endpoints/keys'
 import { updateModel } from '@/api/endpoints/models'
+import { useI18n } from '@/i18n'
 import { parseApiError } from '@/utils/errorParser'
 import type { ProviderWithEndpointsSummary } from '@/api/endpoints'
+import {
+  formatModelMappingEndpointLabel,
+  formatModelMappingRequestScope,
+  modelMappingOperationsKey,
+  normalizeModelMappingOperations,
+} from '../../utils/modelMappingScope'
 import {
   buildDefaultModelTestRequestHeaders,
   buildDefaultModelTestRequestBody,
@@ -424,6 +433,7 @@ const emit = defineEmits<{
 }>()
 
 const { error: showError, success: showSuccess } = useToast()
+const { t } = useI18n()
 
 // 模型测试 composable
 const modelTest = useModelTest({ providerId: () => props.provider.id })
@@ -501,17 +511,46 @@ function getEndpointIdsKey(endpointIds: string[] | undefined): string {
 }
 
 function getOperationsKey(operations: string[] | undefined): string {
-  return getScopeKey(operations)
+  return modelMappingOperationsKey(operations)
 }
 
+const requestScopeLabels = computed(() => ({
+  allRequests: t('providers.modelMapping.scope.allRequests'),
+  sessionCompactionOnly: t('providers.modelMapping.scope.sessionCompactionOnly'),
+  customOperations: (operations: string[]) => t(
+    'providers.modelMapping.scope.customOperations',
+    { operations: operations.join(', ') },
+  ),
+}))
+
 function getGroupEndpointScopeLabel(group: AliasGroup): string {
-  if (!group.endpointIds || group.endpointIds.length === 0) return '全部端点'
-  return `${group.endpointIds.length} 端点`
+  if (!group.endpointIds || group.endpointIds.length === 0) {
+    return t('providers.modelMapping.scope.allEndpoints')
+  }
+  const labels = getGroupEndpointScopeLabels(group)
+  return labels.length === 1
+    ? labels[0]
+    : t('providers.modelMapping.scope.endpointCount', { count: labels.length })
+}
+
+function getGroupEndpointScopeTitle(group: AliasGroup): string {
+  if (!group.endpointIds || group.endpointIds.length === 0) {
+    return t('providers.modelMapping.scope.allEndpoints')
+  }
+  return getGroupEndpointScopeLabels(group).join('、')
+}
+
+function getGroupEndpointScopeLabels(group: AliasGroup): string[] {
+  const endpoints = props.endpoints ?? []
+  return group.endpointIds.map((endpointId) => {
+    const endpoint = endpoints.find(item => item.id === endpointId)
+    if (!endpoint) return endpointId
+    return formatModelMappingEndpointLabel(endpoint, endpoints)
+  })
 }
 
 function getGroupOperationScopeLabel(group: AliasGroup): string {
-  if (group.operations.length === 1 && group.operations[0] === 'compact') return '压缩'
-  return `${group.operations.length} 项操作`
+  return formatModelMappingRequestScope(group.operations, requestScopeLabels.value)
 }
 
 // 精确映射分组（来自 provider_model_mappings）
@@ -536,7 +575,7 @@ const exactMappingGroups = computed<AliasGroup[]>(() => {
           endpointIdsKey,
           endpointIds: normalizeStringList(alias.endpoint_ids),
           operationsKey,
-          operations: normalizeStringList(alias.operations),
+          operations: normalizeModelMappingOperations(alias.operations),
           aliases: []
         }
         groupMap.set(groupKey, group)
diff --git a/frontend/src/features/providers/utils/__tests__/modelMappingScope.spec.ts b/frontend/src/features/providers/utils/__tests__/modelMappingScope.spec.ts
new file mode 100644
index 000000000..bc452acd8
--- /dev/null
+++ b/frontend/src/features/providers/utils/__tests__/modelMappingScope.spec.ts
@@ -0,0 +1,74 @@
+import { describe, expect, it } from 'vitest'
+
+import {
+  ALL_REQUESTS_SCOPE_VALUE,
+  COMPACT_REQUEST_SCOPE_VALUE,
+  formatModelMappingEndpointLabel,
+  formatModelMappingRequestScope,
+  modelMappingOperationsKey,
+  modelMappingOperationsFromScopeValue,
+  modelMappingRequestScopeOptions,
+  modelMappingRequestScopeValue,
+  normalizeModelMappingOperations,
+} from '../modelMappingScope'
+
+describe('model mapping request scope', () => {
+  it('represents an omitted operation filter as all requests', () => {
+    expect(modelMappingRequestScopeValue(undefined)).toBe(ALL_REQUESTS_SCOPE_VALUE)
+    expect(modelMappingOperationsFromScopeValue(ALL_REQUESTS_SCOPE_VALUE)).toBeUndefined()
+    expect(formatModelMappingRequestScope(undefined)).toBe('所有请求')
+  })
+
+  it('round-trips the compact operation as a dedicated request scope', () => {
+    expect(modelMappingRequestScopeValue(['compact'])).toBe(COMPACT_REQUEST_SCOPE_VALUE)
+    expect(modelMappingOperationsFromScopeValue(COMPACT_REQUEST_SCOPE_VALUE)).toEqual(['compact'])
+    expect(formatModelMappingRequestScope(['compact'])).toBe('仅会话压缩')
+  })
+
+  it('normalizes operation values using the backend matching semantics', () => {
+    expect(normalizeModelMappingOperations([' Compact ', 'compact', '', 'SEARCH'])).toEqual([
+      'compact',
+      'search',
+    ])
+    expect(modelMappingOperationsKey(['SEARCH', ' compact ', 'Compact'])).toBe('compact,search')
+  })
+
+  it('preserves an unknown operation scope while editing', () => {
+    const operations = ['future_operation', 'compact']
+    const value = modelMappingRequestScopeValue(operations)
+    const options = modelMappingRequestScopeOptions(operations)
+
+    expect(modelMappingOperationsFromScopeValue(value)).toEqual(operations)
+    expect(options).toContainEqual({ value, label: '仅匹配：future_operation, compact' })
+  })
+
+  it('rejects malformed scope values without constructing operations', () => {
+    expect(modelMappingOperationsFromScopeValue('compact')).toBeUndefined()
+    expect(modelMappingOperationsFromScopeValue('{"compact":true}')).toBeUndefined()
+  })
+
+  it('disambiguates endpoints that share an API format', () => {
+    const endpoints = [
+      {
+        id: 'endpoint-1',
+        api_format: 'openai:responses',
+        base_url: 'https://api.example.com/v1',
+        is_active: true,
+      },
+      {
+        id: 'endpoint-2',
+        api_format: 'openai:responses',
+        base_url: 'https://backup.example.com/v1',
+        custom_path: '/backend-api/codex/responses',
+        is_active: false,
+      },
+    ]
+
+    expect(formatModelMappingEndpointLabel(endpoints[0], endpoints)).toBe(
+      'OpenAI Responses · api.example.com/v1',
+    )
+    expect(formatModelMappingEndpointLabel(endpoints[1], endpoints)).toBe(
+      'OpenAI Responses · backup.example.com/backend-api/codex/responses（停用）',
+    )
+  })
+})
diff --git a/frontend/src/features/providers/utils/modelMappingScope.ts b/frontend/src/features/providers/utils/modelMappingScope.ts
new file mode 100644
index 000000000..248256f31
--- /dev/null
+++ b/frontend/src/features/providers/utils/modelMappingScope.ts
@@ -0,0 +1,132 @@
+import { formatApiFormat } from '@/api/endpoints/types/api-format'
+
+export const MODEL_MAPPING_OPERATION_COMPACT = 'compact'
+
+export const ALL_REQUESTS_SCOPE_VALUE = '[]'
+export const COMPACT_REQUEST_SCOPE_VALUE = JSON.stringify([
+  MODEL_MAPPING_OPERATION_COMPACT,
+])
+
+export interface ModelMappingRequestScopeOption {
+  value: string
+  label: string
+}
+
+export interface ModelMappingRequestScopeLabels {
+  allRequests: string
+  sessionCompactionOnly: string
+  customOperations: (operations: string[]) => string
+}
+
+export interface ModelMappingEndpoint {
+  id: string
+  api_format: string
+  base_url: string
+  custom_path?: string
+  is_active: boolean
+}
+
+const DEFAULT_REQUEST_SCOPE_LABELS: ModelMappingRequestScopeLabels = {
+  allRequests: '所有请求',
+  sessionCompactionOnly: '仅会话压缩',
+  customOperations: operations => `仅匹配：${operations.join(', ')}`,
+}
+
+export function normalizeModelMappingOperations(
+  operations: string[] | undefined,
+): string[] {
+  const seen = new Set<string>()
+  const normalized: string[] = []
+  for (const operation of operations ?? []) {
+    const value = operation.trim().toLowerCase()
+    if (!value || seen.has(value)) continue
+    seen.add(value)
+    normalized.push(value)
+  }
+  return normalized
+}
+
+export function modelMappingRequestScopeValue(
+  operations: string[] | undefined,
+): string {
+  return JSON.stringify(normalizeModelMappingOperations(operations))
+}
+
+export function modelMappingOperationsKey(
+  operations: string[] | undefined,
+): string {
+  return normalizeModelMappingOperations(operations).sort().join(',')
+}
+
+export function modelMappingOperationsFromScopeValue(
+  value: string,
+): string[] | undefined {
+  try {
+    const parsed = JSON.parse(value)
+    if (!Array.isArray(parsed) || parsed.some(item => typeof item !== 'string')) {
+      return undefined
+    }
+    const operations = normalizeModelMappingOperations(parsed)
+    return operations.length > 0 ? operations : undefined
+  } catch {
+    return undefined
+  }
+}
+
+export function formatModelMappingRequestScope(
+  operations: string[] | undefined,
+  labels: ModelMappingRequestScopeLabels = DEFAULT_REQUEST_SCOPE_LABELS,
+): string {
+  const normalized = normalizeModelMappingOperations(operations)
+  if (normalized.length === 0) return labels.allRequests
+  if (
+    normalized.length === 1
+    && normalized[0] === MODEL_MAPPING_OPERATION_COMPACT
+  ) {
+    return labels.sessionCompactionOnly
+  }
+  return labels.customOperations(normalized)
+}
+
+export function modelMappingRequestScopeOptions(
+  operations: string[] | undefined,
+  labels: ModelMappingRequestScopeLabels = DEFAULT_REQUEST_SCOPE_LABELS,
+): ModelMappingRequestScopeOption[] {
+  const options: ModelMappingRequestScopeOption[] = [
+    { value: ALL_REQUESTS_SCOPE_VALUE, label: labels.allRequests },
+    { value: COMPACT_REQUEST_SCOPE_VALUE, label: labels.sessionCompactionOnly },
+  ]
+  const currentValue = modelMappingRequestScopeValue(operations)
+  if (!options.some(option => option.value === currentValue)) {
+    options.push({
+      value: currentValue,
+      label: formatModelMappingRequestScope(operations, labels),
+    })
+  }
+  return options
+}
+
+export function formatModelMappingEndpointLabel(
+  endpoint: ModelMappingEndpoint,
+  endpoints: ModelMappingEndpoint[],
+): string {
+  const sameFormatCount = endpoints.filter(item => item.api_format === endpoint.api_format).length
+  const format = formatApiFormat(endpoint.api_format)
+  const discriminator = sameFormatCount > 1
+    ? formatModelMappingEndpointDiscriminator(endpoint)
+    : ''
+  const status = endpoint.is_active ? '' : '（停用）'
+  return `${format}${discriminator ? ` · ${discriminator}` : ''}${status}`
+}
+
+function formatModelMappingEndpointDiscriminator(endpoint: ModelMappingEndpoint): string {
+  const baseUrl = endpoint.base_url.trim()
+  try {
+    const parsed = new URL(baseUrl)
+    const customPath = endpoint.custom_path?.trim()
+    const path = (customPath || parsed.pathname).replace(/\/$/, '')
+    return `${parsed.host}${path && path !== '/' ? path : ''}`
+  } catch {
+    return baseUrl || endpoint.id.slice(0, 8)
+  }
+}
diff --git a/frontend/src/features/usage/components/UsageRecordsTable.vue b/frontend/src/features/usage/components/UsageRecordsTable.vue
index 83a0a8b37..9426bffa9 100644
--- a/frontend/src/features/usage/components/UsageRecordsTable.vue
+++ b/frontend/src/features/usage/components/UsageRecordsTable.vue
@@ -248,7 +248,7 @@
                 v-if="getRequestTypeLabel(record)"
                 variant="outline"
                 class="h-4 rounded-full border-sky-500/30 bg-sky-500/5 px-1.5 text-[10px] leading-4 text-sky-700 dark:text-sky-300 flex-shrink-0"
-                title="线程压缩"
+                title="会话压缩"
               >
                 {{ getRequestTypeLabel(record) }}
               </Badge>
@@ -765,7 +765,7 @@
                   v-if="getRequestTypeLabel(record)"
                   variant="outline"
                   class="h-4 rounded-full border-sky-500/30 bg-sky-500/5 px-1.5 text-[10px] leading-4 text-sky-700 dark:text-sky-300 flex-shrink-0"
-                  title="线程压缩"
+                  title="会话压缩"
                 >
                   {{ getRequestTypeLabel(record) }}
                 </Badge>
@@ -797,7 +797,7 @@
                 v-if="getRequestTypeLabel(record)"
                 variant="outline"
                 class="h-4 rounded-full border-sky-500/30 bg-sky-500/5 px-1.5 text-[10px] leading-4 text-sky-700 dark:text-sky-300 flex-shrink-0"
-                title="线程压缩"
+                title="会话压缩"
               >
                 {{ getRequestTypeLabel(record) }}
               </Badge>
@@ -1630,7 +1630,7 @@ function getReasoningEffort(record: UsageRecord): string | null {
 }
 
 function getRequestTypeLabel(record: UsageRecord): string | null {
-  return record.request_type?.trim().toLowerCase() === 'compact' ? '压缩' : null
+  return record.request_type?.trim().toLowerCase() === 'compact' ? '会话压缩' : null
 }
 
 function getReasoningEffortTitle(record: UsageRecord): string {
diff --git a/frontend/src/features/usage/components/__tests__/UsageRecordsTable.spec.ts b/frontend/src/features/usage/components/__tests__/UsageRecordsTable.spec.ts
index 27dd0dbee..470935c7e 100644
--- a/frontend/src/features/usage/components/__tests__/UsageRecordsTable.spec.ts
+++ b/frontend/src/features/usage/components/__tests__/UsageRecordsTable.spec.ts
@@ -303,7 +303,7 @@ describe('UsageRecordsTable', () => {
       request_type: 'compact',
     })])
 
-    expect(root.textContent).toContain('压缩')
+    expect(root.textContent).toContain('会话压缩')
   })
 
   it('shows fast badge for priority service tier', () => {
diff --git a/frontend/src/i18n/messages.ts b/frontend/src/i18n/messages.ts
index 156fe8c21..bdedc8b9b 100644
--- a/frontend/src/i18n/messages.ts
+++ b/frontend/src/i18n/messages.ts
@@ -295,6 +295,16 @@ export const messages = {
     'nav.cacheMonitoring': '缓存监控',
     'nav.moduleManagement': '模块管理',
     'nav.systemSettings': '系统设置',
+    'providers.modelMapping.scope.allRequests': '所有请求',
+    'providers.modelMapping.scope.sessionCompactionOnly': '仅会话压缩',
+    'providers.modelMapping.scope.customOperations': '仅匹配：{operations}',
+    'providers.modelMapping.scope.allRequestsDescription': '普通请求和会话压缩都可使用此映射。',
+    'providers.modelMapping.scope.sessionCompactionDescription': '只在会话压缩时使用此映射，包括 Responses 输入中的压缩触发和 Responses Compact 端点。',
+    'providers.modelMapping.scope.customOperationsDescription': '只匹配请求操作：{operations}。',
+    'providers.modelMapping.scope.allEndpoints': '全部端点',
+    'providers.modelMapping.scope.endpointCount': '{count} 个端点',
+    'providers.modelMapping.scope.endpointHelp': '留空时不限制端点；选择后仅匹配指定端点。',
+    'providers.modelMapping.scope.matchHelp': '端点范围和请求范围必须同时匹配。',
     'breadcrumb.personalSettings': '个人设置',
     'breadcrumb.routingCreate': '新建调度策略',
     'breadcrumb.routingConfig': '调度策略配置',
@@ -595,6 +605,16 @@ export const messages = {
     'nav.cacheMonitoring': 'Cache monitoring',
     'nav.moduleManagement': 'Modules',
     'nav.systemSettings': 'System settings',
+    'providers.modelMapping.scope.allRequests': 'All requests',
+    'providers.modelMapping.scope.sessionCompactionOnly': 'Session compaction only',
+    'providers.modelMapping.scope.customOperations': 'Match only: {operations}',
+    'providers.modelMapping.scope.allRequestsDescription': 'Regular requests and session compaction can both use this mapping.',
+    'providers.modelMapping.scope.sessionCompactionDescription': 'Use this mapping only for session compaction, including compaction triggers in Responses input and the Responses Compact endpoint.',
+    'providers.modelMapping.scope.customOperationsDescription': 'Match only these request operations: {operations}.',
+    'providers.modelMapping.scope.allEndpoints': 'All endpoints',
+    'providers.modelMapping.scope.endpointCount': '{count} endpoints',
+    'providers.modelMapping.scope.endpointHelp': 'Leave empty for no endpoint restriction. Selections match only those endpoints.',
+    'providers.modelMapping.scope.matchHelp': 'Both the endpoint scope and request scope must match.',
     'breadcrumb.personalSettings': 'Profile settings',
     'breadcrumb.routingCreate': 'Create routing profile',
     'breadcrumb.routingConfig': 'Routing profile',
@@ -673,6 +693,15 @@ const legacyExactEnglishMessages: Record<string, string> = {
   '全部分组': 'All groups',
   '全部归属': 'All ownership',
   '全部方式': 'All methods',
+  '适用范围': 'Applies to',
+  '适用端点': 'Applicable endpoints',
+  '适用请求': 'Applicable requests',
+  '全部端点': 'All endpoints',
+  '所有请求': 'All requests',
+  '仅会话压缩': 'Session compaction only',
+  '会话压缩': 'Session compaction',
+  '保存映射': 'Save mapping',
+  '添加映射': 'Add mapping',
   '未设置': 'Not set',
   '已设置': 'Configured',
   '默认': 'Default',

From 93e2f95c47942986a52ae56447eb3292fc9eb710 Mon Sep 17 00:00:00 2001
From: MMEXA <youlancelot@gmail.com>
Date: Tue, 14 Jul 2026 02:05:10 +0800
Subject: [PATCH 10/12] =?UTF-8?q?fix(frontend):=20=E6=8C=89=E7=AB=AF?=
 =?UTF-8?q?=E7=82=B9=E8=83=BD=E5=8A=9B=E7=BA=A6=E6=9D=9F=E4=BC=9A=E8=AF=9D?=
 =?UTF-8?q?=E5=8E=8B=E7=BC=A9=E6=98=A0=E5=B0=84?=
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

---
 .../components/ModelMappingDialog.vue         |  39 ++++-
 .../__tests__/ModelMappingDialog.spec.ts      | 147 +++++++++++++++++-
 .../utils/__tests__/modelMappingScope.spec.ts |  41 ++++-
 .../providers/utils/modelMappingScope.ts      |  39 ++++-
 frontend/src/i18n/messages.ts                 |   6 +-
 5 files changed, 262 insertions(+), 10 deletions(-)

diff --git a/frontend/src/features/providers/components/ModelMappingDialog.vue b/frontend/src/features/providers/components/ModelMappingDialog.vue
index 533039941..c224cd361 100644
--- a/frontend/src/features/providers/components/ModelMappingDialog.vue
+++ b/frontend/src/features/providers/components/ModelMappingDialog.vue
@@ -338,6 +338,7 @@ import {
   COMPACT_REQUEST_SCOPE_VALUE,
   formatModelMappingEndpointLabel,
   formatModelMappingRequestScope,
+  modelMappingEndpointScopeSupportsSessionCompaction,
   modelMappingOperationsKey,
   modelMappingOperationsFromScopeValue,
   modelMappingRequestScopeOptions,
@@ -430,6 +431,13 @@ const normalizedSelectedEndpointIds = computed(() => {
   return selected.length > 0 ? selected : undefined
 })
 
+const sessionCompactionScopeAvailable = computed(() => {
+  return modelMappingEndpointScopeSupportsSessionCompaction(
+    normalizedSelectedEndpointIds.value,
+    props.endpoints ?? [],
+  )
+})
+
 const endpointScopeSummary = computed(() => {
   const selected = normalizedSelectedEndpointIds.value
   if (!selected || selected.length === 0) {
@@ -472,12 +480,18 @@ const requestScopeValue = computed(() => {
 })
 
 const requestScopeOptions = computed(() => {
-  return modelMappingRequestScopeOptions(selectedOperations.value, requestScopeLabels.value)
+  return modelMappingRequestScopeOptions(
+    selectedOperations.value,
+    { sessionCompaction: sessionCompactionScopeAvailable.value },
+    requestScopeLabels.value,
+  )
 })
 
 const requestScopeDescription = computed(() => {
   if (requestScopeValue.value === ALL_REQUESTS_SCOPE_VALUE) {
-    return t('providers.modelMapping.scope.allRequestsDescription')
+    return sessionCompactionScopeAvailable.value
+      ? t('providers.modelMapping.scope.allRequestsDescription')
+      : t('providers.modelMapping.scope.allRequestsDefaultDescription')
   }
   if (requestScopeValue.value === COMPACT_REQUEST_SCOPE_VALUE) {
     return t('providers.modelMapping.scope.sessionCompactionDescription')
@@ -487,6 +501,12 @@ const requestScopeDescription = computed(() => {
   })
 })
 
+watch(
+  [() => props.endpoints, () => selectedEndpointIds.value],
+  () => normalizeUnavailableSessionCompactionScope(),
+  { deep: true },
+)
+
 // 所有已知名称集合
 const allKnownNames = computed(() => {
   const set = new Set<string>()
@@ -721,6 +741,7 @@ function initForm() {
   upstreamModels.value = []
   upstreamModelsLoaded.value = false
   collapsedGroups.value = new Set()
+  normalizeUnavailableSessionCompactionScope()
 }
 
 // 处理模型选择变更
@@ -729,9 +750,23 @@ function handleModelChange(value: string) {
 }
 
 function handleRequestScopeChange(value: string) {
+  if (
+    value === COMPACT_REQUEST_SCOPE_VALUE
+    && !sessionCompactionScopeAvailable.value
+  ) return
   selectedOperations.value = modelMappingOperationsFromScopeValue(value) ?? []
 }
 
+function normalizeUnavailableSessionCompactionScope() {
+  if (props.endpoints === undefined) return
+  if (
+    requestScopeValue.value === COMPACT_REQUEST_SCOPE_VALUE
+    && !sessionCompactionScopeAvailable.value
+  ) {
+    selectedOperations.value = []
+  }
+}
+
 // 生成作用域唯一键
 function getApiFormatsKey(formats: string[] | undefined): string {
   return getScopeKey(formats)
diff --git a/frontend/src/features/providers/components/__tests__/ModelMappingDialog.spec.ts b/frontend/src/features/providers/components/__tests__/ModelMappingDialog.spec.ts
index 4885a125c..de095440c 100644
--- a/frontend/src/features/providers/components/__tests__/ModelMappingDialog.spec.ts
+++ b/frontend/src/features/providers/components/__tests__/ModelMappingDialog.spec.ts
@@ -52,8 +52,21 @@ vi.mock('@/components/common/MultiSelect.vue', async () => {
   return {
     default: defineComponent({
       name: 'MultiSelectStub',
-      setup() {
-        return () => h('div')
+      props: {
+        modelValue: { type: Array, default: () => [] },
+        options: { type: Array, default: () => [] },
+      },
+      emits: ['update:modelValue'],
+      setup(props, { emit }) {
+        return () => h('div', (props.options as Array<{ value: string, label: string }>).map(option => h(
+          'button',
+          {
+            type: 'button',
+            'data-endpoint-id': option.value,
+            onClick: () => emit('update:modelValue', [option.value]),
+          },
+          option.label,
+        )))
       },
     }),
   }
@@ -108,6 +121,136 @@ afterEach(() => {
 })
 
 describe('ModelMappingDialog', () => {
+  it('offers session compaction only for an explicitly selected Responses endpoint', async () => {
+    const chatEndpoint = {
+      id: 'endpoint-chat',
+      api_format: 'openai:chat',
+      base_url: 'https://api.example.com/v1',
+      is_active: true,
+    } as ProviderEndpoint
+    const responsesEndpoint = {
+      id: 'endpoint-responses',
+      api_format: 'openai:responses',
+      base_url: 'https://api.example.com/v1',
+      is_active: true,
+    } as ProviderEndpoint
+    const model = {
+      id: 'model-sol',
+      provider_model_name: 'gpt-5.6-sol',
+      global_model_display_name: 'GPT-5.6 Sol',
+      provider_model_mappings: [],
+    } as Model
+    const open = ref(false)
+    const root = document.createElement('div')
+    document.body.appendChild(root)
+    const app = createApp(defineComponent({
+      setup() {
+        return () => h(ModelMappingDialog, {
+          open: open.value,
+          providerId: 'provider-1',
+          endpoints: [chatEndpoint, responsesEndpoint],
+          models: [model],
+          preselectedModelId: model.id,
+          'onUpdate:open': (value: boolean) => { open.value = value },
+        })
+      },
+    }))
+    app.mount(root)
+    mountedApps.push({ app, root })
+
+    open.value = true
+    await nextTick()
+    await nextTick()
+
+    expect(root.textContent).toContain('所有请求')
+    expect(root.textContent).not.toContain('仅会话压缩')
+
+    root.querySelector<HTMLButtonElement>('[data-endpoint-id="endpoint-chat"]')?.click()
+    await nextTick()
+    expect(root.textContent).not.toContain('仅会话压缩')
+
+    root.querySelector<HTMLButtonElement>('[data-endpoint-id="endpoint-responses"]')?.click()
+    await nextTick()
+    expect(root.textContent).toContain('仅会话压缩')
+  })
+
+  it('returns to all requests when a compact mapping switches away from Responses', async () => {
+    const responsesEndpoint = {
+      id: 'endpoint-responses',
+      api_format: 'openai:responses',
+      base_url: 'https://api.example.com/v1',
+      is_active: true,
+    } as ProviderEndpoint
+    const chatEndpoint = {
+      id: 'endpoint-chat',
+      api_format: 'openai:chat',
+      base_url: 'https://api.example.com/v1',
+      is_active: true,
+    } as ProviderEndpoint
+    const model = {
+      id: 'model-sol',
+      provider_model_name: 'gpt-5.6-sol',
+      global_model_display_name: 'GPT-5.6 Sol',
+      provider_model_mappings: [{
+        name: 'gpt-5.6-luna',
+        priority: 1,
+        endpoint_ids: [responsesEndpoint.id],
+        operations: ['compact'],
+      }],
+    } as Model
+    const editingGroup: AliasGroup = {
+      model,
+      apiFormatsKey: '',
+      apiFormats: [],
+      endpointIdsKey: responsesEndpoint.id,
+      endpointIds: [responsesEndpoint.id],
+      operationsKey: 'compact',
+      operations: ['compact'],
+      aliases: model.provider_model_mappings ?? [],
+    }
+    const open = ref(false)
+    const root = document.createElement('div')
+    document.body.appendChild(root)
+    const app = createApp(defineComponent({
+      setup() {
+        return () => h(ModelMappingDialog, {
+          open: open.value,
+          providerId: 'provider-1',
+          endpoints: [responsesEndpoint, chatEndpoint],
+          models: [model],
+          editingGroup,
+          'onUpdate:open': (value: boolean) => { open.value = value },
+        })
+      },
+    }))
+    app.mount(root)
+    mountedApps.push({ app, root })
+
+    open.value = true
+    await nextTick()
+    await nextTick()
+    expect(root.textContent).toContain('仅会话压缩')
+
+    root.querySelector<HTMLButtonElement>('[data-endpoint-id="endpoint-chat"]')?.click()
+    await nextTick()
+    expect(root.textContent).not.toContain('仅会话压缩')
+    expect(root.querySelector<HTMLButtonElement>('[role="radio"][aria-checked="true"]')?.textContent)
+      .toContain('所有请求')
+
+    const saveButton = [...root.querySelectorAll('button')]
+      .find(button => button.textContent?.includes('保存映射'))
+    saveButton?.click()
+    await vi.waitFor(() => expect(updateModel).toHaveBeenCalledTimes(1))
+
+    expect(updateModel).toHaveBeenCalledWith('provider-1', 'model-sol', {
+      provider_model_mappings: [{
+        name: 'gpt-5.6-luna',
+        priority: 1,
+        endpoint_ids: [chatEndpoint.id],
+      }],
+    })
+  })
+
   it('normalizes and replaces an edited compact operation scope', async () => {
     const endpoint = {
       id: 'endpoint-responses',
diff --git a/frontend/src/features/providers/utils/__tests__/modelMappingScope.spec.ts b/frontend/src/features/providers/utils/__tests__/modelMappingScope.spec.ts
index bc452acd8..d43a37f5e 100644
--- a/frontend/src/features/providers/utils/__tests__/modelMappingScope.spec.ts
+++ b/frontend/src/features/providers/utils/__tests__/modelMappingScope.spec.ts
@@ -5,6 +5,7 @@ import {
   COMPACT_REQUEST_SCOPE_VALUE,
   formatModelMappingEndpointLabel,
   formatModelMappingRequestScope,
+  modelMappingEndpointScopeSupportsSessionCompaction,
   modelMappingOperationsKey,
   modelMappingOperationsFromScopeValue,
   modelMappingRequestScopeOptions,
@@ -36,12 +37,50 @@ describe('model mapping request scope', () => {
   it('preserves an unknown operation scope while editing', () => {
     const operations = ['future_operation', 'compact']
     const value = modelMappingRequestScopeValue(operations)
-    const options = modelMappingRequestScopeOptions(operations)
+    const options = modelMappingRequestScopeOptions(
+      operations,
+      { sessionCompaction: true },
+    )
 
     expect(modelMappingOperationsFromScopeValue(value)).toEqual(operations)
     expect(options).toContainEqual({ value, label: '仅匹配：future_operation, compact' })
   })
 
+  it('offers compact scope only when the selected endpoint scope supports it', () => {
+    expect(modelMappingRequestScopeOptions(undefined, { sessionCompaction: false }))
+      .toEqual([{ value: ALL_REQUESTS_SCOPE_VALUE, label: '所有请求' }])
+    expect(modelMappingRequestScopeOptions(undefined, { sessionCompaction: true }))
+      .toContainEqual({ value: COMPACT_REQUEST_SCOPE_VALUE, label: '仅会话压缩' })
+  })
+
+  it('requires every explicitly selected endpoint to use OpenAI Responses', () => {
+    const responsesEndpoint = {
+      id: 'responses',
+      api_format: 'OPENAI_RESPONSES',
+      base_url: 'https://api.example.com/v1',
+      is_active: true,
+    }
+    const chatEndpoint = {
+      id: 'chat',
+      api_format: 'openai:chat',
+      base_url: 'https://api.example.com/v1',
+      is_active: true,
+    }
+
+    expect(modelMappingEndpointScopeSupportsSessionCompaction(
+      undefined,
+      [responsesEndpoint, chatEndpoint],
+    )).toBe(false)
+    expect(modelMappingEndpointScopeSupportsSessionCompaction(
+      [responsesEndpoint.id],
+      [responsesEndpoint, chatEndpoint],
+    )).toBe(true)
+    expect(modelMappingEndpointScopeSupportsSessionCompaction(
+      [responsesEndpoint.id, chatEndpoint.id],
+      [responsesEndpoint, chatEndpoint],
+    )).toBe(false)
+  })
+
   it('rejects malformed scope values without constructing operations', () => {
     expect(modelMappingOperationsFromScopeValue('compact')).toBeUndefined()
     expect(modelMappingOperationsFromScopeValue('{"compact":true}')).toBeUndefined()
diff --git a/frontend/src/features/providers/utils/modelMappingScope.ts b/frontend/src/features/providers/utils/modelMappingScope.ts
index 248256f31..fdac71118 100644
--- a/frontend/src/features/providers/utils/modelMappingScope.ts
+++ b/frontend/src/features/providers/utils/modelMappingScope.ts
@@ -1,4 +1,8 @@
-import { formatApiFormat } from '@/api/endpoints/types/api-format'
+import {
+  API_FORMATS,
+  formatApiFormat,
+  normalizeApiFormatAlias,
+} from '@/api/endpoints/types/api-format'
 
 export const MODEL_MAPPING_OPERATION_COMPACT = 'compact'
 
@@ -12,6 +16,10 @@ export interface ModelMappingRequestScopeOption {
   label: string
 }
 
+export interface ModelMappingRequestScopeCapabilities {
+  sessionCompaction: boolean
+}
+
 export interface ModelMappingRequestScopeLabels {
   allRequests: string
   sessionCompactionOnly: string
@@ -90,14 +98,22 @@ export function formatModelMappingRequestScope(
 
 export function modelMappingRequestScopeOptions(
   operations: string[] | undefined,
+  capabilities: ModelMappingRequestScopeCapabilities,
   labels: ModelMappingRequestScopeLabels = DEFAULT_REQUEST_SCOPE_LABELS,
 ): ModelMappingRequestScopeOption[] {
   const options: ModelMappingRequestScopeOption[] = [
     { value: ALL_REQUESTS_SCOPE_VALUE, label: labels.allRequests },
-    { value: COMPACT_REQUEST_SCOPE_VALUE, label: labels.sessionCompactionOnly },
   ]
+  if (capabilities.sessionCompaction) {
+    options.push({
+      value: COMPACT_REQUEST_SCOPE_VALUE,
+      label: labels.sessionCompactionOnly,
+    })
+  }
   const currentValue = modelMappingRequestScopeValue(operations)
-  if (!options.some(option => option.value === currentValue)) {
+  const compactScopeUnavailable = currentValue === COMPACT_REQUEST_SCOPE_VALUE
+    && !capabilities.sessionCompaction
+  if (!compactScopeUnavailable && !options.some(option => option.value === currentValue)) {
     options.push({
       value: currentValue,
       label: formatModelMappingRequestScope(operations, labels),
@@ -106,6 +122,23 @@ export function modelMappingRequestScopeOptions(
   return options
 }
 
+export function modelMappingEndpointScopeSupportsSessionCompaction(
+  endpointIds: string[] | undefined,
+  endpoints: ModelMappingEndpoint[],
+): boolean {
+  const selectedIds = [...new Set(
+    (endpointIds ?? []).map(id => id.trim()).filter(Boolean),
+  )]
+  if (selectedIds.length === 0) return false
+
+  const endpointsById = new Map(endpoints.map(endpoint => [endpoint.id, endpoint]))
+  return selectedIds.every((endpointId) => {
+    const endpoint = endpointsById.get(endpointId)
+    return endpoint !== undefined
+      && normalizeApiFormatAlias(endpoint.api_format) === API_FORMATS.OPENAI_RESPONSES
+  })
+}
+
 export function formatModelMappingEndpointLabel(
   endpoint: ModelMappingEndpoint,
   endpoints: ModelMappingEndpoint[],
diff --git a/frontend/src/i18n/messages.ts b/frontend/src/i18n/messages.ts
index bdedc8b9b..725fefa76 100644
--- a/frontend/src/i18n/messages.ts
+++ b/frontend/src/i18n/messages.ts
@@ -299,7 +299,8 @@ export const messages = {
     'providers.modelMapping.scope.sessionCompactionOnly': '仅会话压缩',
     'providers.modelMapping.scope.customOperations': '仅匹配：{operations}',
     'providers.modelMapping.scope.allRequestsDescription': '普通请求和会话压缩都可使用此映射。',
-    'providers.modelMapping.scope.sessionCompactionDescription': '只在会话压缩时使用此映射，包括 Responses 输入中的压缩触发和 Responses Compact 端点。',
+    'providers.modelMapping.scope.allRequestsDefaultDescription': '所选端点的所有请求都可使用此映射。仅当适用端点已明确选择且全部为 OpenAI Responses 时，可限定为会话压缩。',
+    'providers.modelMapping.scope.sessionCompactionDescription': '仅匹配所选 OpenAI Responses 端点中的会话压缩请求。',
     'providers.modelMapping.scope.customOperationsDescription': '只匹配请求操作：{operations}。',
     'providers.modelMapping.scope.allEndpoints': '全部端点',
     'providers.modelMapping.scope.endpointCount': '{count} 个端点',
@@ -609,7 +610,8 @@ export const messages = {
     'providers.modelMapping.scope.sessionCompactionOnly': 'Session compaction only',
     'providers.modelMapping.scope.customOperations': 'Match only: {operations}',
     'providers.modelMapping.scope.allRequestsDescription': 'Regular requests and session compaction can both use this mapping.',
-    'providers.modelMapping.scope.sessionCompactionDescription': 'Use this mapping only for session compaction, including compaction triggers in Responses input and the Responses Compact endpoint.',
+    'providers.modelMapping.scope.allRequestsDefaultDescription': 'This mapping applies to all requests on the selected endpoints. Session compaction can be selected only when every explicitly selected endpoint uses OpenAI Responses.',
+    'providers.modelMapping.scope.sessionCompactionDescription': 'Match only session compaction requests on the selected OpenAI Responses endpoints.',
     'providers.modelMapping.scope.customOperationsDescription': 'Match only these request operations: {operations}.',
     'providers.modelMapping.scope.allEndpoints': 'All endpoints',
     'providers.modelMapping.scope.endpointCount': '{count} endpoints',

From f9c343eb07472ca7928376d1bb4aa19c65c0d178 Mon Sep 17 00:00:00 2001
From: MMEXA <youlancelot@gmail.com>
Date: Sat, 18 Jul 2026 05:55:06 +0800
Subject: [PATCH 11/12] =?UTF-8?q?fix(gateway):=20=E9=80=82=E9=85=8D=20Rust?=
 =?UTF-8?q?=201.95=20=E6=95=B4=E9=99=A4=E6=A3=80=E6=9F=A5?=
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

---
 apps/aether-gateway/src/handlers/shared/catalog.rs | 6 +++---
 1 file changed, 3 insertions(+), 3 deletions(-)

diff --git a/apps/aether-gateway/src/handlers/shared/catalog.rs b/apps/aether-gateway/src/handlers/shared/catalog.rs
index 94fbcea51..98b098c1f 100644
--- a/apps/aether-gateway/src/handlers/shared/catalog.rs
+++ b/apps/aether-gateway/src/handlers/shared/catalog.rs
@@ -821,11 +821,11 @@ fn codex_quota_period_identity(window_minutes: u64) -> (String, String) {
         return ("monthly".to_string(), "月".to_string());
     }
 
-    let label = if window_minutes % MINUTES_PER_WEEK == 0 {
+    let label = if window_minutes.is_multiple_of(MINUTES_PER_WEEK) {
         format!("{}周", window_minutes / MINUTES_PER_WEEK)
-    } else if window_minutes % MINUTES_PER_DAY == 0 {
+    } else if window_minutes.is_multiple_of(MINUTES_PER_DAY) {
         format!("{}天", window_minutes / MINUTES_PER_DAY)
-    } else if window_minutes % MINUTES_PER_HOUR == 0 {
+    } else if window_minutes.is_multiple_of(MINUTES_PER_HOUR) {
         format!("{}H", window_minutes / MINUTES_PER_HOUR)
     } else {
         format!("{window_minutes}分钟")

From ac3796af84526b8aa12846f513dcab946a801961 Mon Sep 17 00:00:00 2001
From: MMEXA <youlancelot@gmail.com>
Date: Sat, 18 Jul 2026 06:31:40 +0800
Subject: [PATCH 12/12] =?UTF-8?q?fix(gateway):=20=E6=81=A2=E5=A4=8D?=
 =?UTF-8?q?=E5=93=8D=E5=BA=94=E8=BE=B9=E7=95=8C=E5=B9=B6=E7=BB=9F=E4=B8=80?=
 =?UTF-8?q?=E6=A0=BC=E5=BC=8F=E5=85=A5=E5=8F=A3?=
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

---
 .../stream/execution_failures.rs               |  2 +-
 .../src/execution_runtime/transport.rs         | 18 +++++++++---------
 2 files changed, 10 insertions(+), 10 deletions(-)

diff --git a/apps/aether-gateway/src/execution_runtime/stream/execution_failures.rs b/apps/aether-gateway/src/execution_runtime/stream/execution_failures.rs
index 97885eb86..5207d705d 100644
--- a/apps/aether-gateway/src/execution_runtime/stream/execution_failures.rs
+++ b/apps/aether-gateway/src/execution_runtime/stream/execution_failures.rs
@@ -522,7 +522,7 @@ mod tests {
             "data: {\"type\":\"response.failed\",\"response\":{\"status\":\"failed\",\"error\":{\"type\":\"invalid_request\",\"message\":\"This content was flagged for possible cybersecurity risk.\",\"code\":\"cyber_policy_violation\",\"param\":\"input\",\"details\":{\"policy_category\":\"cybersecurity\",\"appeal_allowed\":true}}}}\n\n",
         )
         .as_bytes();
-        let terminal_error = aether_ai_formats::api::extract_provider_private_stream_error_body(
+        let terminal_error = crate::ai_serving::api::extract_provider_private_stream_error_body(
             None,
             provider_buffered_body,
         )
diff --git a/apps/aether-gateway/src/execution_runtime/transport.rs b/apps/aether-gateway/src/execution_runtime/transport.rs
index 231e33923..4cf035b11 100644
--- a/apps/aether-gateway/src/execution_runtime/transport.rs
+++ b/apps/aether-gateway/src/execution_runtime/transport.rs
@@ -3760,6 +3760,15 @@ pub(crate) fn build_execution_response_body(
         return Ok(None);
     }
 
+    if !stream && response_body_is_json(headers, decoded_body_bytes) {
+        let body_json: Value = serde_json::from_slice(decoded_body_bytes)
+            .map_err(ExecutionRuntimeTransportError::InvalidJson)?;
+        return Ok(Some(ResponseBody {
+            json_body: Some(body_json),
+            body_bytes_b64: None,
+        }));
+    }
+
     if let Some(body_json) = extract_provider_private_stream_error_body(None, decoded_body_bytes)
         .or_else(|| extract_provider_private_stream_error_body(None, body_bytes))
     {
@@ -3776,15 +3785,6 @@ pub(crate) fn build_execution_response_body(
         }));
     }
 
-    if response_body_is_json(headers, decoded_body_bytes) {
-        let body_json: Value = serde_json::from_slice(decoded_body_bytes)
-            .map_err(ExecutionRuntimeTransportError::InvalidJson)?;
-        return Ok(Some(ResponseBody {
-            json_body: Some(body_json),
-            body_bytes_b64: None,
-        }));
-    }
-
     Ok(Some(ResponseBody {
         json_body: None,
         body_bytes_b64: Some(base64::engine::general_purpose::STANDARD.encode(body_bytes)),