fix(gateway): use Vertex model garden catalog endpoint

crates/aether-model-fetch/src/strategy.rs

@@ -464,8 +464,6 @@ async fn fetch_vertex_service_account_models(
         });
     };
     let token = exchange_vertex_service_account_token(runtime, &transports[0], auth_config).await?;
-    let project_id = json_string(auth_config.get("project_id"))
-        .ok_or_else(|| "vertex_ai(service_account): missing project_id".to_string())?;
     let gemini_transport =
         select_transport_for_api_format(transports, "gemini:").unwrap_or(&transports[0]);
     let claude_transport =
@@ -476,18 +474,12 @@ async fn fetch_vertex_service_account_models(
     let mut soft_errors = Vec::new();
     let mut has_success = false;
 
-    for region in vertex_regions(auth_config) {
+    for base in iter_vertex_base_urls(transports) {
-        let base = if region == "global" {
-            VERTEX_API_BASE_URL.to_string()
-        } else {
-            format!("https://{region}-aiplatform.googleapis.com")
-        };
         for (publisher, transport, api_format) in [
             ("google", gemini_transport, "gemini:generate_content"),
             ("anthropic", claude_transport, "claude:messages"),
         ] {
-            let url =
+            let url = build_vertex_service_account_list_url(&base, publisher, None);
-                build_vertex_service_account_list_url(&base, &project_id, &region, publisher, None);
             let outcome = fetch_vertex_models_from_url(
                 runtime,
                 transport,
@@ -929,14 +921,7 @@ fn iter_vertex_base_urls(transports: &[GatewayProviderTransportSnapshot]) -> Vec
 }
 
 fn build_vertex_google_list_url(base_url: &str, api_key: &str, page_token: Option<&str>) -> String {
-    let path = if base_url.trim_end_matches('/').ends_with("/v1")
+    let url = build_vertex_publisher_models_list_base_url(base_url, "google");
-        || base_url.trim_end_matches('/').ends_with("/v1beta")
-    {
-        "/publishers/google/models"
-    } else {
-        "/v1/publishers/google/models"
-    };
-    let url = build_simple_path_url(base_url, path);
     let mut url = append_query_param(url, "key", api_key);
     url = append_query_param(url, "pageSize", VERTEX_PAGE_SIZE);
     if let Some(page_token) = page_token {
@@ -947,14 +932,10 @@ fn build_vertex_google_list_url(base_url: &str, api_key: &str, page_token: Optio
 
 fn build_vertex_service_account_list_url(
     base_url: &str,
-    project_id: &str,
-    region: &str,
     publisher: &str,
     page_token: Option<&str>,
 ) -> String {
-    let path =
+    let mut url = build_vertex_publisher_models_list_base_url(base_url, publisher);
-        format!("/v1/projects/{project_id}/locations/{region}/publishers/{publisher}/models");
-    let mut url = build_simple_path_url(base_url, &path);
     url = append_query_param(url, "pageSize", VERTEX_PAGE_SIZE);
     if let Some(page_token) = page_token {
         url = append_query_param(url, "pageToken", page_token);
@@ -962,6 +943,16 @@ fn build_vertex_service_account_list_url(
     url
 }
 
+fn build_vertex_publisher_models_list_base_url(base_url: &str, publisher: &str) -> String {
+    let trimmed_base = base_url.trim().trim_end_matches('/');
+    let path = if trimmed_base.ends_with("/v1") || trimmed_base.ends_with("/v1beta1") {
+        format!("/publishers/{publisher}/models")
+    } else {
+        format!("/v1beta1/publishers/{publisher}/models")
+    };
+    build_simple_path_url(trimmed_base, &path)
+}
+
 fn build_simple_path_url(base_url: &str, path: &str) -> String {
     format!("{}{}", base_url.trim().trim_end_matches('/'), path.trim())
 }
@@ -1087,38 +1078,6 @@ fn vertex_effective_format(model_id: &str, auth_config: Option<&Value>) -> Strin
     }
 }
 
-fn vertex_regions(auth_config: &Value) -> Vec<String> {
-    let mut seen = BTreeSet::new();
-    let mut regions = Vec::new();
-    let auth_config = auth_config.as_object();
-
-    let mut push_region = |region: Option<&str>| {
-        let Some(region) = region.map(str::trim).filter(|value| !value.is_empty()) else {
-            return;
-        };
-        if seen.insert(region.to_string()) {
-            regions.push(region.to_string());
-        }
-    };
-
-    push_region(
-        auth_config
-            .and_then(|value| value.get("region"))
-            .and_then(Value::as_str),
-    );
-    if let Some(model_regions) = auth_config
-        .and_then(|value| value.get("model_regions"))
-        .and_then(Value::as_object)
-    {
-        for value in model_regions.values() {
-            push_region(value.as_str());
-        }
-    }
-    push_region(Some("global"));
-    push_region(Some("us-central1"));
-    regions
-}
-
 fn is_soft_not_found(error: &str) -> bool {
     error.trim().starts_with("HTTP 404:")
 }
@@ -1498,7 +1457,7 @@ mod tests {
         let urls = executed_urls.lock().expect("executed_urls lock");
         assert_eq!(
             urls.as_slice(),
-            &["https://aiplatform.googleapis.com/v1/publishers/google/models?key=vertex-secret&pageSize=100"]
+            &["https://aiplatform.googleapis.com/v1beta1/publishers/google/models?key=vertex-secret&pageSize=100"]
         );
         assert_eq!(outcome.fetched_model_ids, vec!["gemini-3.1-pro-preview"]);
         assert_eq!(outcome.cached_models.len(), 1);
@@ -1508,6 +1467,24 @@ mod tests {
         );
     }
 
+    #[test]
+    fn vertex_publisher_models_list_url_uses_model_garden_resource_not_runtime_resource() {
+        let url = super::build_vertex_service_account_list_url(
+            "https://aiplatform.googleapis.com",
+            "google",
+            None,
+        );
+
+        assert_eq!(
+            url,
+            "https://aiplatform.googleapis.com/v1beta1/publishers/google/models?pageSize=100"
+        );
+        assert!(
+            !url.contains("/projects/") && !url.contains("/locations/"),
+            "Model Garden publisher list must not use Vertex runtime project/location path"
+        );
+    }
+
     #[tokio::test]
     async fn codex_transport_fetches_upstream_models_instead_of_preset_catalog() {
         let executed_urls = Arc::new(Mutex::new(Vec::new()));

docs/architecture/gemini-api-endpoint-routing.md

@@ -79,6 +79,7 @@ Vertex AI 的 Gemini API REST reference 使用 `aiplatform.googleapis.com` 或 r
 - Vertex AI Embed Content REST: <https://docs.cloud.google.com/vertex-ai/generative-ai/docs/reference/rest/v1/projects.locations.publishers.models/embedContent>
 - Vertex AI Predict REST: <https://docs.cloud.google.com/vertex-ai/generative-ai/docs/reference/rest/v1/projects.locations.publishers.models/predict>
 - Vertex AI REST resources: <https://docs.cloud.google.com/vertex-ai/generative-ai/docs/reference/rest/v1/projects.locations.publishers.models>
+- Vertex AI Model Garden publisher model list: <https://docs.cloud.google.com/vertex-ai/docs/reference/rest/v1beta1/publishers.models/list>
 - Vertex AI text embeddings API: <https://cloud.google.com/vertex-ai/generative-ai/docs/model-reference/text-embeddings-api>
 - Vertex AI OpenAI compatibility: <https://cloud.google.com/vertex-ai/generative-ai/docs/start/openai>
 
@@ -89,6 +90,9 @@ Vertex AI 的 Gemini API REST reference 使用 `aiplatform.googleapis.com` 或 r
   - 对 `global` location，可使用 `https://aiplatform.googleapis.com/v1/projects/{project}/locations/global/...`
 - Vertex API key 路径可走：
   - `https://aiplatform.googleapis.com/v1/publishers/google/models/{model}:{action}?key=...`
+- Vertex 模型目录拉取不是推理请求，必须走 Model Garden publisher list：
+  - `https://aiplatform.googleapis.com/v1beta1/publishers/{publisher}/models`
+  - 不得使用 `projects/{project}/locations/{location}/publishers/{publisher}/models`；`projects.locations.publishers.models` 资源没有 list 方法，只有 generate / stream / predict / embed 等动作。
 - Vertex 文本 embedding API 文档使用 `:predict`，请求体是 `instances[]`，可选参数在 `parameters` 下；响应是 `predictions[].embeddings.values`。
 - Vertex REST reference 也列出 `embedContent`，但 Aether 当前 text embedding 主链使用 text embeddings guide 和 Predict API 的 contract。
 - Vertex `instances[]` 是在线 Predict 请求体，不等同于异步 batch prediction job。模型级输入数量限制由 Vertex 返回；Aether 不把超出限制的请求静默改走其他产品面。