diff --git a/components/backend/cmd/sync_flags.go b/components/backend/cmd/sync_flags.go
index e96e83edf..6d453d196 100644
--- a/components/backend/cmd/sync_flags.go
+++ b/components/backend/cmd/sync_flags.go
@@ -48,11 +48,17 @@ type FlagsConfig struct {
}
// FlagsFromManifest converts a model manifest into FlagSpecs.
-// Skips the default model and unavailable models.
+// Skips default models (global and per-provider) and unavailable models.
func FlagsFromManifest(manifest *types.ModelManifest) []FlagSpec {
+ // Build set of all default model IDs (global + per-provider)
+ defaults := map[string]bool{manifest.DefaultModel: true}
+ for _, id := range manifest.ProviderDefaults {
+ defaults[id] = true
+ }
+
var specs []FlagSpec
for _, model := range manifest.Models {
- if model.ID == manifest.DefaultModel {
+ if defaults[model.ID] {
continue
}
if !model.Available {
diff --git a/components/backend/cmd/sync_flags_test.go b/components/backend/cmd/sync_flags_test.go
index fa92238d1..2d2ddb61c 100644
--- a/components/backend/cmd/sync_flags_test.go
+++ b/components/backend/cmd/sync_flags_test.go
@@ -66,22 +66,44 @@ func TestParseManifestPath(t *testing.T) {
func TestFlagsFromManifest_SkipsDefaultAndUnavailable(t *testing.T) {
manifest := &types.ModelManifest{
DefaultModel: "claude-sonnet-4-5",
+ ProviderDefaults: map[string]string{
+ "anthropic": "claude-sonnet-4-5",
+ "google": "gemini-2.5-flash",
+ },
Models: []types.ModelEntry{
- {ID: "claude-sonnet-4-5", Label: "Sonnet 4.5", Available: true},
- {ID: "claude-opus-4-6", Label: "Opus 4.6", Available: true},
- {ID: "claude-opus-4-1", Label: "Opus 4.1", Available: false},
+ {ID: "claude-sonnet-4-5", Label: "Sonnet 4.5", Provider: "anthropic", Available: true},
+ {ID: "claude-opus-4-6", Label: "Opus 4.6", Provider: "anthropic", Available: true},
+ {ID: "claude-opus-4-1", Label: "Opus 4.1", Provider: "anthropic", Available: false},
+ {ID: "gemini-2.5-flash", Label: "Gemini 2.5 Flash", Provider: "google", Available: true},
+ {ID: "gemini-2.5-pro", Label: "Gemini 2.5 Pro", Provider: "google", Available: true},
},
}
flags := FlagsFromManifest(manifest)
- if len(flags) != 1 {
- t.Fatalf("expected 1 flag, got %d: %v", len(flags), flags)
+
+ // Should skip: claude-sonnet-4-5 (global default + anthropic default),
+ // gemini-2.5-flash (google default),
+ // claude-opus-4-1 (unavailable)
+ // Should include: claude-opus-4-6, gemini-2.5-pro
+ if len(flags) != 2 {
+ t.Fatalf("expected 2 flags, got %d: %v", len(flags), flags)
+ }
+
+ names := map[string]bool{}
+ for _, f := range flags {
+ names[f.Name] = true
+ }
+ if !names["model.claude-opus-4-6.enabled"] {
+ t.Error("expected model.claude-opus-4-6.enabled")
+ }
+ if !names["model.gemini-2.5-pro.enabled"] {
+ t.Error("expected model.gemini-2.5-pro.enabled")
}
- if flags[0].Name != "model.claude-opus-4-6.enabled" {
- t.Errorf("expected model.claude-opus-4-6.enabled, got %s", flags[0].Name)
+ if names["model.claude-sonnet-4-5.enabled"] {
+ t.Error("global default should be skipped")
}
- if len(flags[0].Tags) != 1 || flags[0].Tags[0].Type != "scope" || flags[0].Tags[0].Value != "workspace" {
- t.Errorf("expected scope:workspace tag, got %v", flags[0].Tags)
+ if names["model.gemini-2.5-flash.enabled"] {
+ t.Error("provider default should be skipped")
}
}
diff --git a/components/backend/handlers/models.go b/components/backend/handlers/models.go
index d6bb15ab4..a8d40e24c 100644
--- a/components/backend/handlers/models.go
+++ b/components/backend/handlers/models.go
@@ -54,14 +54,15 @@ func ListModelsForProject(c *gin.Context) {
ctx := c.Request.Context()
namespace := sanitizeParam(c.Param("projectName"))
+ providerFilter := sanitizeParam(c.Query("provider"))
manifest, err := LoadManifest(ManifestPath())
if err != nil {
- log.Printf("WARNING: failed to load model manifest: %v", err)
+ log.Printf("WARNING: failed to load model manifest from disk: %v", err)
manifest = cachedManifest.Load()
if manifest == nil {
- log.Printf("WARNING: no cached manifest available, using hardcoded defaults")
- c.JSON(http.StatusOK, defaultModelsResponse())
+ log.Printf("ERROR: no model manifest available (file unreadable, no cache)")
+ c.JSON(http.StatusServiceUnavailable, gin.H{"error": "Model manifest unavailable"})
return
}
} else {
@@ -76,13 +77,27 @@ func ListModelsForProject(c *gin.Context) {
// Continue without overrides
}
- var models []types.Model
+ // Resolve which model ID is the "default" for this request.
+ // When filtering by provider, use the provider-specific default.
+ effectiveDefault := manifest.DefaultModel
+ if providerFilter != "" {
+ if pd, ok := manifest.ProviderDefaults[providerFilter]; ok {
+ effectiveDefault = pd
+ }
+ }
+
+ models := make([]types.Model, 0)
for _, entry := range manifest.Models {
if !entry.Available {
continue
}
- isDefault := entry.ID == manifest.DefaultModel
+ // Filter by provider if specified
+ if providerFilter != "" && entry.Provider != providerFilter {
+ continue
+ }
+
+ isDefault := entry.ID == effectiveDefault
flagName := fmt.Sprintf("model.%s.enabled", entry.ID)
// Default model is always included
@@ -103,15 +118,15 @@ func ListModelsForProject(c *gin.Context) {
}
}
+ responseDefault := effectiveDefault
if len(models) == 0 {
- log.Printf("WARNING: no models passed filtering, using defaults")
- c.JSON(http.StatusOK, defaultModelsResponse())
- return
+ log.Printf("WARNING: no models passed filtering for provider=%q in namespace %s", providerFilter, namespace)
+ responseDefault = ""
}
c.JSON(http.StatusOK, types.ListModelsResponse{
Models: models,
- DefaultModel: manifest.DefaultModel,
+ DefaultModel: responseDefault,
})
}
@@ -143,94 +158,68 @@ func LoadManifest(path string) (*types.ModelManifest, error) {
return &manifest, nil
}
-// isModelAvailable checks if a model is available for session creation.
-//
-// Validation strategy:
-// 1. Check the agent registry — if the model is declared in the selected
-// runner's model list, it's valid. This is the primary check for all runners.
-// 2. For models also in the models.json manifest (Claude models), additionally
-// check feature-flag gating and workspace overrides.
-// 3. If the model is not found in either source, reject it.
-func isModelAvailable(ctx context.Context, k8sClient kubernetes.Interface, modelID, runnerTypeID, namespace string) bool {
+// isModelAvailable checks if a model is available for session creation in the
+// given workspace namespace. All models (Claude and Gemini) are validated
+// against models.json. Returns true if the model exists, is available, and
+// is enabled (checking workspace overrides first, then Unleash).
+// When requiredProvider is non-empty, the model's provider must match
+// (prevents using a Gemini model with a Claude runner, for example).
+// The default model always returns true. Fails open when no manifest has
+// ever been loaded (cold start).
+func isModelAvailable(ctx context.Context, k8sClient kubernetes.Interface, modelID, requiredProvider, namespace string) bool {
if modelID == "" {
return true // Empty model will use default
}
- // 1. Check agent registry — runner-specific model validation
- rt, err := GetRuntime(runnerTypeID)
- if err == nil && len(rt.Models) > 0 {
- found := false
- for _, m := range rt.Models {
- if m.Value == modelID {
- found = true
- break
- }
- }
- if !found {
- log.Printf("Model %q not in runner %q model list, rejecting", modelID, runnerTypeID)
- return false
- }
- // Model is in the runner's list — now check if it also needs
- // feature-flag gating via the manifest (applies to Claude models).
- }
-
- // 2. Check models.json manifest for feature-flag gating (if applicable)
manifest, err := LoadManifest(ManifestPath())
if err != nil {
- log.Printf("WARNING: failed to load model manifest: %v", err)
+ log.Printf("WARNING: failed to load model manifest for validation: %v", err)
manifest = cachedManifest.Load()
+ if manifest == nil {
+ // When we know the runner's provider, reject unknown models rather
+ // than allowing a cross-provider mismatch through to the runner.
+ // Fail-open only when both manifest and registry are unavailable
+ // (requiredProvider == "") to avoid blocking cold starts.
+ if requiredProvider != "" {
+ log.Printf("WARNING: no manifest available, rejecting model %q (provider=%q)", modelID, requiredProvider)
+ return false
+ }
+ log.Printf("WARNING: no manifest or registry available, allowing model %q", modelID)
+ return true
+ }
} else {
cachedManifest.Store(manifest)
}
- if manifest != nil {
- // Default model is always available
- if modelID == manifest.DefaultModel {
- return true
- }
- for _, entry := range manifest.Models {
- if entry.ID == modelID {
- if !entry.Available {
- return false
- }
- flagName := fmt.Sprintf("model.%s.enabled", entry.ID)
- overrides, oErr := getWorkspaceOverrides(ctx, k8sClient, namespace)
- if oErr != nil {
- log.Printf("WARNING: failed to read workspace overrides for %s: %v", namespace, oErr)
+ for _, entry := range manifest.Models {
+ if entry.ID == modelID {
+ if !entry.Available {
+ return false
+ }
+ // Provider mismatch check applies to ALL models, including defaults
+ if requiredProvider != "" && entry.Provider != requiredProvider {
+ log.Printf("Model %q has provider %q but runner requires %q", modelID, entry.Provider, requiredProvider)
+ return false
+ }
+ // Default models (global and per-provider) are always enabled
+ // (skip feature flag check) but must still pass provider matching above
+ if modelID == manifest.DefaultModel {
+ return true
+ }
+ for provider, pd := range manifest.ProviderDefaults {
+ if modelID == pd && (requiredProvider == "" || provider == requiredProvider) {
+ return true
}
- return isModelEnabledWithOverrides(flagName, overrides)
}
+ flagName := fmt.Sprintf("model.%s.enabled", entry.ID)
+ overrides, oErr := getWorkspaceOverrides(ctx, k8sClient, namespace)
+ if oErr != nil {
+ log.Printf("WARNING: failed to read workspace overrides for %s: %v", namespace, oErr)
+ }
+ return isModelEnabledWithOverrides(flagName, overrides)
}
}
- // 3. If we validated via registry in step 1 (found=true), allow it.
- // Models not in the manifest skip feature-flag gating (e.g., Gemini models).
- if rt != nil && len(rt.Models) > 0 {
- return true // Already validated in step 1
- }
-
- // No manifest loaded and no registry available — fail-open on cold start
- if manifest == nil {
- log.Printf("WARNING: no manifest or registry available, allowing model %q", modelID)
- return true
- }
-
- log.Printf("WARNING: model %q not found in manifest or agent registry, rejecting", modelID)
+ log.Printf("WARNING: model %q not found in manifest, rejecting", modelID)
return false
}
-
-// defaultModelsResponse returns a hardcoded ListModelsResponse as a fallback
-// when the model manifest file is unavailable or malformed.
-// Keep in sync with components/manifests/base/models.json (available: true entries).
-func defaultModelsResponse() types.ListModelsResponse {
- return types.ListModelsResponse{
- DefaultModel: "claude-sonnet-4-5",
- Models: []types.Model{
- {ID: "claude-sonnet-4-5", Label: "Claude Sonnet 4.5", Provider: "anthropic", IsDefault: true},
- {ID: "claude-sonnet-4-6", Label: "Claude Sonnet 4.6", Provider: "anthropic", IsDefault: false},
- {ID: "claude-opus-4-6", Label: "Claude Opus 4.6", Provider: "anthropic", IsDefault: false},
- {ID: "claude-opus-4-5", Label: "Claude Opus 4.5", Provider: "anthropic", IsDefault: false},
- {ID: "claude-haiku-4-5", Label: "Claude Haiku 4.5", Provider: "anthropic", IsDefault: false},
- },
- }
-}
diff --git a/components/backend/handlers/models_test.go b/components/backend/handlers/models_test.go
index 65e1076b3..7a6eb7576 100644
--- a/components/backend/handlers/models_test.go
+++ b/components/backend/handlers/models_test.go
@@ -36,13 +36,19 @@ var _ = Describe("Models Handler", Label(test_constants.LabelUnit, test_constant
)
validManifestObj := types.ModelManifest{
- Version: 1,
+ Version: 2,
DefaultModel: "claude-sonnet-4-5",
+ ProviderDefaults: map[string]string{
+ "anthropic": "claude-sonnet-4-5",
+ "google": "gemini-2.5-flash",
+ },
Models: []types.ModelEntry{
{ID: "claude-sonnet-4-5", Label: "Claude Sonnet 4.5", VertexID: "claude-sonnet-4-5@20250929", Provider: "anthropic", Available: true},
{ID: "claude-opus-4-6", Label: "Claude Opus 4.6", VertexID: "claude-opus-4-6@default", Provider: "anthropic", Available: true},
{ID: "claude-opus-4-5", Label: "Claude Opus 4.5", VertexID: "claude-opus-4-5@20251101", Provider: "anthropic", Available: true},
{ID: "claude-haiku-4-5", Label: "Claude Haiku 4.5", VertexID: "claude-haiku-4-5@20251001", Provider: "anthropic", Available: true},
+ {ID: "gemini-2.5-flash", Label: "Gemini 2.5 Flash", VertexID: "gemini-2.5-flash", Provider: "google", Available: true},
+ {ID: "gemini-2.5-pro", Label: "Gemini 2.5 Pro", VertexID: "gemini-2.5-pro", Provider: "google", Available: true},
},
}
@@ -130,8 +136,8 @@ var _ = Describe("Models Handler", Label(test_constants.LabelUnit, test_constant
var resp types.ListModelsResponse
err := json.Unmarshal(httpTestUtils.GetResponseRecorder().Body.Bytes(), &resp)
Expect(err).NotTo(HaveOccurred())
- // With no Unleash configured, IsModelEnabled returns true, so all 4 models pass
- Expect(resp.Models).To(HaveLen(4))
+ // With no Unleash configured, IsModelEnabled returns true, so all 6 models pass
+ Expect(resp.Models).To(HaveLen(6))
Expect(resp.DefaultModel).To(Equal("claude-sonnet-4-5"))
})
@@ -224,9 +230,9 @@ var _ = Describe("Models Handler", Label(test_constants.LabelUnit, test_constant
err := json.Unmarshal(httpTestUtils.GetResponseRecorder().Body.Bytes(), &resp)
Expect(err).NotTo(HaveOccurred())
- // opus-4-6 excluded by override; the other 3 should still be present
- // (default model + 2 non-default models via Unleash fallback which returns true when not configured)
- Expect(resp.Models).To(HaveLen(3))
+ // opus-4-6 excluded by override; the other 5 should still be present
+ // (default model + 4 non-default models via Unleash fallback which returns true when not configured)
+ Expect(resp.Models).To(HaveLen(5))
ids := make([]string, len(resp.Models))
for i, m := range resp.Models {
ids[i] = m.ID
@@ -234,6 +240,8 @@ var _ = Describe("Models Handler", Label(test_constants.LabelUnit, test_constant
Expect(ids).To(ContainElement("claude-sonnet-4-5"))
Expect(ids).To(ContainElement("claude-opus-4-5"))
Expect(ids).To(ContainElement("claude-haiku-4-5"))
+ Expect(ids).To(ContainElement("gemini-2.5-flash"))
+ Expect(ids).To(ContainElement("gemini-2.5-pro"))
Expect(ids).NotTo(ContainElement("claude-opus-4-6"))
})
@@ -284,15 +292,15 @@ var _ = Describe("Models Handler", Label(test_constants.LabelUnit, test_constant
var resp types.ListModelsResponse
err = json.Unmarshal(httpTestUtils.GetResponseRecorder().Body.Bytes(), &resp)
Expect(err).NotTo(HaveOccurred())
- Expect(resp.Models).To(HaveLen(3))
+ Expect(resp.Models).To(HaveLen(5))
for _, m := range resp.Models {
Expect(m.ID).NotTo(Equal("claude-opus-4-6"))
}
})
- It("should return hardcoded defaults when manifest file is missing and no cache", func() {
- logger.Log("Testing ListModelsForProject fallback when manifest file missing and no cache")
+ It("should return 503 when manifest file is missing and no cache", func() {
+ logger.Log("Testing ListModelsForProject returns 503 when manifest unavailable")
os.Setenv("MODELS_MANIFEST_PATH", filepath.Join(GinkgoT().TempDir(), "nonexistent.json"))
fakeClient := setupK8sWithOverrides()
setupAuth(fakeClient)
@@ -300,13 +308,7 @@ var _ = Describe("Models Handler", Label(test_constants.LabelUnit, test_constant
ginCtx := createAuthenticatedContext("test-project")
ListModelsForProject(ginCtx)
- httpTestUtils.AssertHTTPStatus(http.StatusOK)
-
- var resp types.ListModelsResponse
- err := json.Unmarshal(httpTestUtils.GetResponseRecorder().Body.Bytes(), &resp)
- Expect(err).NotTo(HaveOccurred())
- Expect(resp.Models).To(HaveLen(5))
- Expect(resp.DefaultModel).To(Equal("claude-sonnet-4-5"))
+ httpTestUtils.AssertHTTPStatus(http.StatusServiceUnavailable)
})
It("should use cached manifest when file becomes unavailable", func() {
@@ -333,14 +335,13 @@ var _ = Describe("Models Handler", Label(test_constants.LabelUnit, test_constant
var resp types.ListModelsResponse
err := json.Unmarshal(httpTestUtils.GetResponseRecorder().Body.Bytes(), &resp)
Expect(err).NotTo(HaveOccurred())
- // Cached manifest has 4 models (not 5 hardcoded defaults),
- // and they go through flag filtering
- Expect(resp.Models).To(HaveLen(4))
+ // Cached manifest has 6 models and they go through flag filtering
+ Expect(resp.Models).To(HaveLen(6))
Expect(resp.DefaultModel).To(Equal("claude-sonnet-4-5"))
})
- It("should return hardcoded defaults when JSON is malformed and no cache", func() {
- logger.Log("Testing ListModelsForProject fallback with malformed JSON and no cache")
+ It("should return 503 when JSON is malformed and no cache", func() {
+ logger.Log("Testing ListModelsForProject returns 503 with malformed JSON and no cache")
writeManifestFile("{invalid json")
fakeClient := setupK8sWithOverrides()
setupAuth(fakeClient)
@@ -348,14 +349,79 @@ var _ = Describe("Models Handler", Label(test_constants.LabelUnit, test_constant
ginCtx := createAuthenticatedContext("test-project")
ListModelsForProject(ginCtx)
+ httpTestUtils.AssertHTTPStatus(http.StatusServiceUnavailable)
+ })
+ })
+
+ Context("ListModelsForProject with provider filter", func() {
+ It("should return only anthropic models when provider=anthropic", func() {
+ logger.Log("Testing provider filter for anthropic")
+ writeManifestFile(validManifest)
+ fakeClient := setupK8sWithOverrides()
+ setupAuth(fakeClient)
+
+ ginCtx := httpTestUtils.CreateTestGinContext("GET", "/api/projects/test-project/models?provider=anthropic", nil)
+ httpTestUtils.SetAuthHeader("test-token")
+ ginCtx.Params = gin.Params{{Key: "projectName", Value: "test-project"}}
+ ginCtx.Request.URL.RawQuery = "provider=anthropic"
+
+ ListModelsForProject(ginCtx)
+
httpTestUtils.AssertHTTPStatus(http.StatusOK)
var resp types.ListModelsResponse
err := json.Unmarshal(httpTestUtils.GetResponseRecorder().Body.Bytes(), &resp)
Expect(err).NotTo(HaveOccurred())
- Expect(resp.Models).To(HaveLen(5))
+
+ for _, m := range resp.Models {
+ Expect(m.Provider).To(Equal("anthropic"), "All models should be anthropic")
+ }
+ Expect(resp.Models).To(HaveLen(4))
Expect(resp.DefaultModel).To(Equal("claude-sonnet-4-5"))
})
+
+ It("should return only google models when provider=google", func() {
+ logger.Log("Testing provider filter for google")
+ writeManifestFile(validManifest)
+ fakeClient := setupK8sWithOverrides()
+ setupAuth(fakeClient)
+
+ ginCtx := httpTestUtils.CreateTestGinContext("GET", "/api/projects/test-project/models?provider=google", nil)
+ httpTestUtils.SetAuthHeader("test-token")
+ ginCtx.Params = gin.Params{{Key: "projectName", Value: "test-project"}}
+ ginCtx.Request.URL.RawQuery = "provider=google"
+
+ ListModelsForProject(ginCtx)
+
+ httpTestUtils.AssertHTTPStatus(http.StatusOK)
+
+ var resp types.ListModelsResponse
+ err := json.Unmarshal(httpTestUtils.GetResponseRecorder().Body.Bytes(), &resp)
+ Expect(err).NotTo(HaveOccurred())
+
+ for _, m := range resp.Models {
+ Expect(m.Provider).To(Equal("google"), "All models should be google")
+ }
+ Expect(resp.Models).To(HaveLen(2))
+ Expect(resp.DefaultModel).To(Equal("gemini-2.5-flash"))
+ })
+
+ It("should return all models when no provider filter", func() {
+ logger.Log("Testing no provider filter returns all models")
+ writeManifestFile(validManifest)
+ fakeClient := setupK8sWithOverrides()
+ setupAuth(fakeClient)
+
+ ginCtx := createAuthenticatedContext("test-project")
+ ListModelsForProject(ginCtx)
+
+ httpTestUtils.AssertHTTPStatus(http.StatusOK)
+
+ var resp types.ListModelsResponse
+ err := json.Unmarshal(httpTestUtils.GetResponseRecorder().Body.Bytes(), &resp)
+ Expect(err).NotTo(HaveOccurred())
+ Expect(resp.Models).To(HaveLen(6))
+ })
})
Context("isModelEnabledWithOverrides", func() {
@@ -392,9 +458,11 @@ var _ = Describe("Models Handler", Label(test_constants.LabelUnit, test_constant
manifest, err := LoadManifest(path)
Expect(err).NotTo(HaveOccurred())
- Expect(manifest.Version).To(Equal(1))
+ Expect(manifest.Version).To(Equal(2))
Expect(manifest.DefaultModel).To(Equal("claude-sonnet-4-5"))
- Expect(manifest.Models).To(HaveLen(4))
+ Expect(manifest.ProviderDefaults).To(HaveLen(2))
+ Expect(manifest.ProviderDefaults["google"]).To(Equal("gemini-2.5-flash"))
+ Expect(manifest.Models).To(HaveLen(6))
})
It("should return error when file is missing", func() {
@@ -415,7 +483,7 @@ var _ = Describe("Models Handler", Label(test_constants.LabelUnit, test_constant
Context("isModelAvailable", func() {
It("should return true for empty model ID", func() {
logger.Log("Testing isModelAvailable with empty model ID")
- result := isModelAvailable(context.Background(), K8sClient, "", "claude-agent-sdk", "test-ns")
+ result := isModelAvailable(context.Background(), K8sClient, "", "", "test-ns")
Expect(result).To(BeTrue())
})
@@ -424,7 +492,7 @@ var _ = Describe("Models Handler", Label(test_constants.LabelUnit, test_constant
writeManifestFile(validManifest)
setupK8sWithOverrides()
- result := isModelAvailable(context.Background(), K8sClient, "claude-sonnet-4-5", "claude-agent-sdk", "test-ns")
+ result := isModelAvailable(context.Background(), K8sClient, "claude-sonnet-4-5", "", "test-ns")
Expect(result).To(BeTrue())
})
@@ -433,7 +501,7 @@ var _ = Describe("Models Handler", Label(test_constants.LabelUnit, test_constant
writeManifestFile(validManifest)
setupK8sWithOverrides()
- result := isModelAvailable(context.Background(), K8sClient, "claude-opus-4-6", "claude-agent-sdk", "test-ns")
+ result := isModelAvailable(context.Background(), K8sClient, "claude-opus-4-6", "", "test-ns")
Expect(result).To(BeTrue())
})
@@ -450,7 +518,7 @@ var _ = Describe("Models Handler", Label(test_constants.LabelUnit, test_constant
writeManifestFile(string(manifestBytes))
setupK8sWithOverrides()
- result := isModelAvailable(context.Background(), K8sClient, "claude-opus-4-6", "claude-agent-sdk", "test-ns")
+ result := isModelAvailable(context.Background(), K8sClient, "claude-opus-4-6", "", "test-ns")
Expect(result).To(BeFalse())
})
@@ -459,18 +527,26 @@ var _ = Describe("Models Handler", Label(test_constants.LabelUnit, test_constant
writeManifestFile(validManifest)
setupK8sWithOverrides()
- result := isModelAvailable(context.Background(), K8sClient, "nonexistent-model", "claude-agent-sdk", "test-ns")
+ result := isModelAvailable(context.Background(), K8sClient, "nonexistent-model", "", "test-ns")
Expect(result).To(BeFalse())
})
- It("should fail-open when manifest file is missing", func() {
- logger.Log("Testing isModelAvailable fail-open when manifest file missing")
+ It("should fail-open when manifest is missing and no provider required", func() {
+ logger.Log("Testing isModelAvailable fail-open when manifest missing and requiredProvider empty")
os.Setenv("MODELS_MANIFEST_PATH", filepath.Join(GinkgoT().TempDir(), "nonexistent.json"))
- result := isModelAvailable(context.Background(), K8sClient, "claude-opus-4-6", "claude-agent-sdk", "test-ns")
+ result := isModelAvailable(context.Background(), K8sClient, "claude-opus-4-6", "", "test-ns")
Expect(result).To(BeTrue())
})
+ It("should reject when manifest is missing but provider is required", func() {
+ logger.Log("Testing isModelAvailable rejects when manifest missing and requiredProvider set")
+ os.Setenv("MODELS_MANIFEST_PATH", filepath.Join(GinkgoT().TempDir(), "nonexistent.json"))
+
+ result := isModelAvailable(context.Background(), K8sClient, "claude-opus-4-6", "anthropic", "test-ns")
+ Expect(result).To(BeFalse())
+ })
+
It("should return false when workspace override disables the model", func() {
logger.Log("Testing isModelAvailable respects workspace override=false")
writeManifestFile(validManifest)
@@ -485,7 +561,7 @@ var _ = Describe("Models Handler", Label(test_constants.LabelUnit, test_constant
}
setupK8sWithOverrides(overrideCM)
- result := isModelAvailable(context.Background(), K8sClient, "claude-opus-4-6", "claude-agent-sdk", "test-project")
+ result := isModelAvailable(context.Background(), K8sClient, "claude-opus-4-6", "", "test-project")
Expect(result).To(BeFalse())
})
@@ -503,7 +579,35 @@ var _ = Describe("Models Handler", Label(test_constants.LabelUnit, test_constant
}
setupK8sWithOverrides(overrideCM)
- result := isModelAvailable(context.Background(), K8sClient, "claude-opus-4-6", "claude-agent-sdk", "test-project")
+ result := isModelAvailable(context.Background(), K8sClient, "claude-opus-4-6", "", "test-project")
+ Expect(result).To(BeTrue())
+ })
+
+ It("should reject provider-default model when provider does not match requiredProvider", func() {
+ logger.Log("Testing isModelAvailable rejects provider-default with wrong provider")
+ writeManifestFile(validManifest)
+ setupK8sWithOverrides()
+
+ // gemini-2.5-flash is the google provider default — should be rejected for anthropic runner
+ result := isModelAvailable(context.Background(), K8sClient, "gemini-2.5-flash", "anthropic", "test-ns")
+ Expect(result).To(BeFalse())
+ })
+
+ It("should reject model when provider does not match requiredProvider", func() {
+ logger.Log("Testing isModelAvailable rejects provider mismatch")
+ writeManifestFile(validManifest)
+ setupK8sWithOverrides()
+
+ result := isModelAvailable(context.Background(), K8sClient, "claude-opus-4-6", "google", "test-ns")
+ Expect(result).To(BeFalse())
+ })
+
+ It("should accept model when provider matches requiredProvider", func() {
+ logger.Log("Testing isModelAvailable accepts matching provider")
+ writeManifestFile(validManifest)
+ setupK8sWithOverrides()
+
+ result := isModelAvailable(context.Background(), K8sClient, "claude-opus-4-6", "anthropic", "test-ns")
Expect(result).To(BeTrue())
})
})
diff --git a/components/backend/handlers/runner_types.go b/components/backend/handlers/runner_types.go
index 626167cf4..c4875918a 100644
--- a/components/backend/handlers/runner_types.go
+++ b/components/backend/handlers/runner_types.go
@@ -23,16 +23,15 @@ const DefaultRunnerPort = 8001
// Keep both in sync when modifying the schema.
// It is the single source of truth for runtime configuration.
type AgentRuntimeSpec struct {
- ID string `json:"id"`
- DisplayName string `json:"displayName"`
- Description string `json:"description"`
- Framework string `json:"framework"`
- Container ContainerSpec `json:"container"`
- Sandbox SandboxSpec `json:"sandbox"`
- Auth AuthSpec `json:"auth"`
- DefaultModel string `json:"defaultModel"`
- Models []ModelOption `json:"models"`
- FeatureGate string `json:"featureGate"`
+ ID string `json:"id"`
+ DisplayName string `json:"displayName"`
+ Description string `json:"description"`
+ Framework string `json:"framework"`
+ Container ContainerSpec `json:"container"`
+ Sandbox SandboxSpec `json:"sandbox"`
+ Auth AuthSpec `json:"auth"`
+ Provider string `json:"provider"`
+ FeatureGate string `json:"featureGate"`
}
// ContainerSpec defines the runner container configuration.
@@ -72,23 +71,16 @@ type AuthSpec struct {
VertexSupported bool `json:"vertexSupported"`
}
-// ModelOption represents a model choice within a runner type.
-type ModelOption struct {
- Value string `json:"value"`
- Label string `json:"label"`
-}
-
// RunnerTypeResponse is the public API shape returned to the frontend.
// FeatureGate is intentionally excluded — gated runners are already filtered
// out by the handler, so the frontend never needs to see the gate name.
type RunnerTypeResponse struct {
- ID string `json:"id"`
- DisplayName string `json:"displayName"`
- Description string `json:"description"`
- Framework string `json:"framework"`
- DefaultModel string `json:"defaultModel"`
- Models []ModelOption `json:"models"`
- Auth AuthSpec `json:"auth"`
+ ID string `json:"id"`
+ DisplayName string `json:"displayName"`
+ Description string `json:"description"`
+ Framework string `json:"framework"`
+ Provider string `json:"provider"`
+ Auth AuthSpec `json:"auth"`
}
// In-memory cache for the agent registry (ConfigMap content changes rarely).
@@ -221,9 +213,20 @@ func isRunnerEnabled(runnerID string) bool {
return FeatureEnabled(rt.FeatureGate)
}
-// GetRunnerTypes handles GET /api/runner-types and returns the list of available runner types.
-// Runners gated by feature flags are filtered out.
-func GetRunnerTypes(c *gin.Context) {
+// isRunnerEnabledWithOverrides checks workspace ConfigMap overrides first,
+// then falls back to the Unleash SDK for global state.
+func isRunnerEnabledWithOverrides(flagName string, overrides map[string]string) bool {
+ if overrides != nil {
+ if val, exists := overrides[flagName]; exists {
+ return val == "true"
+ }
+ }
+ return FeatureEnabled(flagName)
+}
+
+// GetRunnerTypesGlobal handles GET /api/runner-types (no auth, no workspace overrides).
+// Used by admin pages that need to list all runner types regardless of workspace.
+func GetRunnerTypesGlobal(c *gin.Context) {
entries, err := loadAgentRegistry()
if err != nil {
log.Printf("Failed to load agent registry: %v", err)
@@ -233,19 +236,61 @@ func GetRunnerTypes(c *gin.Context) {
resp := make([]RunnerTypeResponse, 0, len(entries))
for _, e := range entries {
- // Check feature gate directly instead of calling isRunnerEnabled (which
- // re-loads the registry per entry — N+1 pattern).
if e.FeatureGate != "" && !FeatureEnabled(e.FeatureGate) {
continue
}
resp = append(resp, RunnerTypeResponse{
- ID: e.ID,
- DisplayName: e.DisplayName,
- Description: e.Description,
- Framework: e.Framework,
- DefaultModel: e.DefaultModel,
- Models: e.Models,
- Auth: e.Auth,
+ ID: e.ID,
+ DisplayName: e.DisplayName,
+ Description: e.Description,
+ Framework: e.Framework,
+ Provider: e.Provider,
+ Auth: e.Auth,
+ })
+ }
+
+ c.JSON(http.StatusOK, resp)
+}
+
+// GetRunnerTypes handles GET /api/projects/:projectName/runner-types and returns
+// the list of available runner types. Runners gated by feature flags are filtered
+// out, respecting workspace-scoped overrides in the feature-flag-overrides ConfigMap.
+func GetRunnerTypes(c *gin.Context) {
+ reqK8s, _ := GetK8sClientsForRequest(c)
+ if reqK8s == nil {
+ c.JSON(http.StatusUnauthorized, gin.H{"error": "User token required"})
+ c.Abort()
+ return
+ }
+
+ ctx := c.Request.Context()
+ namespace := sanitizeParam(c.Param("projectName"))
+
+ entries, err := loadAgentRegistry()
+ if err != nil {
+ log.Printf("Failed to load agent registry: %v", err)
+ c.JSON(http.StatusInternalServerError, gin.H{"error": "Failed to load runner types"})
+ return
+ }
+
+ // Load workspace overrides for feature gate evaluation
+ overrides, err := getWorkspaceOverrides(ctx, reqK8s, namespace)
+ if err != nil {
+ log.Printf("WARNING: failed to read workspace overrides for runner types in %s: %v", namespace, err)
+ }
+
+ resp := make([]RunnerTypeResponse, 0, len(entries))
+ for _, e := range entries {
+ if e.FeatureGate != "" && !isRunnerEnabledWithOverrides(e.FeatureGate, overrides) {
+ continue
+ }
+ resp = append(resp, RunnerTypeResponse{
+ ID: e.ID,
+ DisplayName: e.DisplayName,
+ Description: e.Description,
+ Framework: e.Framework,
+ Provider: e.Provider,
+ Auth: e.Auth,
})
}
diff --git a/components/backend/handlers/runner_types_test.go b/components/backend/handlers/runner_types_test.go
index 32b10e682..ba6b91380 100644
--- a/components/backend/handlers/runner_types_test.go
+++ b/components/backend/handlers/runner_types_test.go
@@ -10,6 +10,9 @@ import (
"time"
"github.com/gin-gonic/gin"
+ "k8s.io/apimachinery/pkg/runtime"
+ dynamicfake "k8s.io/client-go/dynamic/fake"
+ "k8s.io/client-go/kubernetes/fake"
)
// sampleRegistryJSON returns a test agent registry JSON with 2 runtimes.
@@ -20,6 +23,7 @@ func sampleRegistryJSON() string {
DisplayName: "Claude Code",
Description: "Anthropic Claude with full coding capabilities",
Framework: "claude-agent-sdk",
+ Provider: "anthropic",
Container: ContainerSpec{
Image: "quay.io/ambient_code/ambient_runner:latest",
Port: 8001,
@@ -42,11 +46,6 @@ func sampleRegistryJSON() string {
SecretKeyLogic: "any",
VertexSupported: true,
},
- DefaultModel: "claude-sonnet-4-5",
- Models: []ModelOption{
- {Value: "claude-sonnet-4-5", Label: "Claude Sonnet 4.5"},
- {Value: "claude-opus-4-6", Label: "Claude Opus 4.6"},
- },
FeatureGate: "",
},
{
@@ -54,6 +53,7 @@ func sampleRegistryJSON() string {
DisplayName: "Gemini CLI",
Description: "Google Gemini coding agent",
Framework: "gemini-cli",
+ Provider: "google",
Container: ContainerSpec{
Image: "quay.io/ambient_code/ambient_runner:latest",
Port: 9090,
@@ -72,10 +72,6 @@ func sampleRegistryJSON() string {
SecretKeyLogic: "any",
VertexSupported: true,
},
- DefaultModel: "gemini-2.5-flash",
- Models: []ModelOption{
- {Value: "gemini-2.5-flash", Label: "Gemini 2.5 Flash"},
- },
FeatureGate: "runner.gemini-cli.enabled",
},
}
@@ -88,13 +84,21 @@ func sampleRegistryJSON() string {
func setupRegistryForTest(t *testing.T) {
t.Helper()
- // Write test registry JSON to a temp file and point AGENT_REGISTRY_PATH at it
- tmpDir := t.TempDir()
- tmpFile := filepath.Join(tmpDir, "agent-registry.json")
- if err := os.WriteFile(tmpFile, []byte(sampleRegistryJSON()), 0644); err != nil {
+ // Write registry JSON to a temp file and point env var to it
+ dir := t.TempDir()
+ path := filepath.Join(dir, "agent-registry.json")
+ if err := os.WriteFile(path, []byte(sampleRegistryJSON()), 0644); err != nil {
t.Fatalf("Failed to write test registry: %v", err)
}
- t.Setenv("AGENT_REGISTRY_PATH", tmpFile)
+ t.Setenv("AGENT_REGISTRY_PATH", path)
+
+ // Set up fake K8s clients for auth and workspace overrides
+ K8sClientMw = fake.NewSimpleClientset()
+ DynamicClient = dynamicfake.NewSimpleDynamicClient(runtime.NewScheme())
+
+ if Namespace == "" {
+ Namespace = "test-ns"
+ }
// Clear the in-memory cache
registryCacheMu.Lock()
@@ -146,12 +150,12 @@ func TestGetRuntime_FullFields(t *testing.T) {
t.Fatalf("GetRuntime failed: %v", err)
}
- // Framework
if rt.Framework != "claude-agent-sdk" {
t.Errorf("Framework: expected 'claude-agent-sdk', got %q", rt.Framework)
}
-
- // Auth
+ if rt.Provider != "anthropic" {
+ t.Errorf("Provider: expected 'anthropic', got %q", rt.Provider)
+ }
if len(rt.Auth.RequiredSecretKeys) != 1 || rt.Auth.RequiredSecretKeys[0] != "ANTHROPIC_API_KEY" {
t.Errorf("Auth.RequiredSecretKeys: expected [ANTHROPIC_API_KEY], got %v", rt.Auth.RequiredSecretKeys)
}
@@ -161,19 +165,9 @@ func TestGetRuntime_FullFields(t *testing.T) {
if !rt.Auth.VertexSupported {
t.Error("Auth.VertexSupported: expected true")
}
-
- // FeatureGate
if rt.FeatureGate != "" {
t.Errorf("FeatureGate: expected empty string, got %q", rt.FeatureGate)
}
-
- // Models
- if len(rt.Models) != 2 {
- t.Errorf("Expected 2 models, got %d", len(rt.Models))
- }
- if rt.DefaultModel != "claude-sonnet-4-5" {
- t.Errorf("DefaultModel: expected 'claude-sonnet-4-5', got %q", rt.DefaultModel)
- }
}
func TestGetRuntime_GeminiFields(t *testing.T) {
@@ -187,6 +181,9 @@ func TestGetRuntime_GeminiFields(t *testing.T) {
if rt.Framework != "gemini-cli" {
t.Errorf("Framework: expected 'gemini-cli', got %q", rt.Framework)
}
+ if rt.Provider != "google" {
+ t.Errorf("Provider: expected 'google', got %q", rt.Provider)
+ }
if rt.FeatureGate != "runner.gemini-cli.enabled" {
t.Errorf("FeatureGate: expected 'runner.gemini-cli.enabled', got %q", rt.FeatureGate)
}
@@ -272,18 +269,16 @@ func TestGetContainerEnvVars_UnknownFallback(t *testing.T) {
// --- GetRunnerTypes handler test ---
-func TestGetRunnerTypes_ReturnsFullFields(t *testing.T) {
+func TestGetRunnerTypes_ReturnsProvider(t *testing.T) {
setupRegistryForTest(t)
- // Without Unleash initialized, FeatureEnabled returns false.
- // isRunnerEnabled returns true for runtimes with empty featureGate,
- // and false for runtimes with a non-empty featureGate.
- // In our test data: claude (featureGate="") -> enabled, gemini (featureGate="runner.gemini-cli.enabled") -> disabled.
-
gin.SetMode(gin.TestMode)
w := httptest.NewRecorder()
c, _ := gin.CreateTestContext(w)
- c.Request = httptest.NewRequest(http.MethodGet, "/api/runner-types", nil)
+ req := httptest.NewRequest(http.MethodGet, "/api/projects/test-project/runner-types", nil)
+ req.Header.Set("Authorization", "Bearer test-token")
+ c.Request = req
+ c.Params = gin.Params{{Key: "projectName", Value: "test-project"}}
GetRunnerTypes(c)
@@ -297,7 +292,6 @@ func TestGetRunnerTypes_ReturnsFullFields(t *testing.T) {
}
// Only claude-agent-sdk should be returned (empty featureGate = always enabled)
- // gemini-cli has featureGate="runner.gemini-cli.enabled" which is disabled without Unleash
if len(resp) != 1 {
t.Fatalf("Expected 1 runner type (only ungated), got %d", len(resp))
}
@@ -306,36 +300,27 @@ func TestGetRunnerTypes_ReturnsFullFields(t *testing.T) {
if claude.ID != "claude-agent-sdk" {
t.Fatalf("Expected claude-agent-sdk, got %q", claude.ID)
}
-
- // Verify full AgentRuntimeSpec fields are in the response
+ if claude.Provider != "anthropic" {
+ t.Errorf("Provider: expected 'anthropic', got %q", claude.Provider)
+ }
if claude.Framework != "claude-agent-sdk" {
t.Errorf("Framework: expected 'claude-agent-sdk', got %q", claude.Framework)
}
if claude.Auth.SecretKeyLogic != "any" {
t.Errorf("Auth.SecretKeyLogic: expected 'any', got %q", claude.Auth.SecretKeyLogic)
}
- if claude.Auth.VertexSupported != true {
- t.Error("Auth.VertexSupported: expected true")
- }
- if len(claude.Auth.RequiredSecretKeys) != 1 || claude.Auth.RequiredSecretKeys[0] != "ANTHROPIC_API_KEY" {
- t.Errorf("Auth.RequiredSecretKeys: expected [ANTHROPIC_API_KEY], got %v", claude.Auth.RequiredSecretKeys)
- }
- if claude.DefaultModel != "claude-sonnet-4-5" {
- t.Errorf("DefaultModel: expected 'claude-sonnet-4-5', got %q", claude.DefaultModel)
- }
- if len(claude.Models) != 2 {
- t.Errorf("Expected 2 models, got %d", len(claude.Models))
- }
}
func TestGetRunnerTypes_GatedRunnersFiltered(t *testing.T) {
setupRegistryForTest(t)
- // Without Unleash, gated runners should be filtered out
gin.SetMode(gin.TestMode)
w := httptest.NewRecorder()
c, _ := gin.CreateTestContext(w)
- c.Request = httptest.NewRequest(http.MethodGet, "/api/runner-types", nil)
+ req := httptest.NewRequest(http.MethodGet, "/api/projects/test-project/runner-types", nil)
+ req.Header.Set("Authorization", "Bearer test-token")
+ c.Request = req
+ c.Params = gin.Params{{Key: "projectName", Value: "test-project"}}
GetRunnerTypes(c)
@@ -352,7 +337,6 @@ func TestGetRunnerTypes_GatedRunnersFiltered(t *testing.T) {
func TestIsRunnerEnabled_EmptyGate(t *testing.T) {
setupRegistryForTest(t)
- // Runtimes with empty featureGate should always be enabled
if !isRunnerEnabled("claude-agent-sdk") {
t.Error("claude-agent-sdk with empty featureGate should be enabled")
}
@@ -361,8 +345,88 @@ func TestIsRunnerEnabled_EmptyGate(t *testing.T) {
func TestIsRunnerEnabled_NonEmptyGate_Disabled(t *testing.T) {
setupRegistryForTest(t)
- // Without Unleash, non-empty featureGate should be disabled
if isRunnerEnabled("gemini-cli") {
t.Error("gemini-cli should be disabled when Unleash is not configured")
}
}
+
+// --- isRunnerEnabledWithOverrides tests ---
+
+func TestIsRunnerEnabledWithOverrides_OverrideTrue(t *testing.T) {
+ overrides := map[string]string{"runner.gemini-cli.enabled": "true"}
+ if !isRunnerEnabledWithOverrides("runner.gemini-cli.enabled", overrides) {
+ t.Error("expected enabled when override is true")
+ }
+}
+
+func TestIsRunnerEnabledWithOverrides_OverrideFalse(t *testing.T) {
+ overrides := map[string]string{"runner.gemini-cli.enabled": "false"}
+ if isRunnerEnabledWithOverrides("runner.gemini-cli.enabled", overrides) {
+ t.Error("expected disabled when override is false")
+ }
+}
+
+func TestIsRunnerEnabledWithOverrides_NoOverrideFallsThrough(t *testing.T) {
+ overrides := map[string]string{"other.flag": "true"}
+ // Without Unleash configured, FeatureEnabled returns false
+ if isRunnerEnabledWithOverrides("runner.gemini-cli.enabled", overrides) {
+ t.Error("expected disabled when no override and Unleash not configured")
+ }
+}
+
+func TestIsRunnerEnabledWithOverrides_NilOverrides(t *testing.T) {
+ // Without Unleash configured, FeatureEnabled returns false
+ if isRunnerEnabledWithOverrides("runner.gemini-cli.enabled", nil) {
+ t.Error("expected disabled with nil overrides and Unleash not configured")
+ }
+}
+
+// --- GetRunnerTypesGlobal tests ---
+
+func TestGetRunnerTypesGlobal_ReturnsUngatedRunners(t *testing.T) {
+ setupRegistryForTest(t)
+
+ gin.SetMode(gin.TestMode)
+ w := httptest.NewRecorder()
+ c, _ := gin.CreateTestContext(w)
+ c.Request = httptest.NewRequest(http.MethodGet, "/api/runner-types", nil)
+
+ GetRunnerTypesGlobal(c)
+
+ if w.Code != http.StatusOK {
+ t.Fatalf("Expected 200, got %d: %s", w.Code, w.Body.String())
+ }
+
+ var resp []RunnerTypeResponse
+ if err := json.Unmarshal(w.Body.Bytes(), &resp); err != nil {
+ t.Fatalf("Failed to parse response: %v", err)
+ }
+
+ // Only ungated runners returned (gemini-cli gated, disabled without Unleash)
+ if len(resp) != 1 {
+ t.Fatalf("Expected 1 runner type, got %d", len(resp))
+ }
+ if resp[0].ID != "claude-agent-sdk" {
+ t.Errorf("Expected claude-agent-sdk, got %q", resp[0].ID)
+ }
+ if resp[0].Provider != "anthropic" {
+ t.Errorf("Expected provider anthropic, got %q", resp[0].Provider)
+ }
+}
+
+func TestGetRunnerTypesGlobal_NoAuthRequired(t *testing.T) {
+ setupRegistryForTest(t)
+
+ gin.SetMode(gin.TestMode)
+ w := httptest.NewRecorder()
+ c, _ := gin.CreateTestContext(w)
+ // No auth header
+ c.Request = httptest.NewRequest(http.MethodGet, "/api/runner-types", nil)
+
+ GetRunnerTypesGlobal(c)
+
+ // Should succeed without auth
+ if w.Code != http.StatusOK {
+ t.Fatalf("Expected 200 without auth, got %d", w.Code)
+ }
+}
diff --git a/components/backend/handlers/sessions.go b/components/backend/handlers/sessions.go
index 19e1071ed..aa93a53fb 100644
--- a/components/backend/handlers/sessions.go
+++ b/components/backend/handlers/sessions.go
@@ -624,8 +624,18 @@ func CreateSession(c *gin.Context) {
}
}
- // Validate that the requested model is available for this runner type
- if llmSettings.Model != "" && !isModelAvailable(c.Request.Context(), reqK8s, llmSettings.Model, runnerTypeID, project) {
+ // Validate model availability with provider matching.
+ // If the runner type is found in the registry, enforce that the model's
+ // provider matches the runner's provider. If the registry is unavailable
+ // (e.g., ConfigMap not mounted), skip provider matching but still validate
+ // the model against the manifest.
+ runnerProvider := ""
+ if rt, rtErr := GetRuntime(runnerTypeID); rtErr == nil {
+ runnerProvider = rt.Provider
+ } else {
+ log.Printf("WARNING: could not resolve runner type %q from registry: %v", runnerTypeID, rtErr)
+ }
+ if llmSettings.Model != "" && !isModelAvailable(c.Request.Context(), reqK8s, llmSettings.Model, runnerProvider, project) {
c.JSON(http.StatusBadRequest, gin.H{"error": "Model is not available for this runner type"})
return
}
diff --git a/components/backend/routes.go b/components/backend/routes.go
index 177d8e4db..955e4fe57 100644
--- a/components/backend/routes.go
+++ b/components/backend/routes.go
@@ -13,13 +13,15 @@ func registerRoutes(r *gin.Engine) {
{
// Public endpoints (no auth required)
api.GET("/workflows/ootb", handlers.ListOOTBWorkflows)
- api.GET("/runner-types", handlers.GetRunnerTypes)
+ // Global runner-types endpoint (no workspace overrides — for admin pages)
+ api.GET("/runner-types", handlers.GetRunnerTypesGlobal)
api.POST("/projects/:projectName/agentic-sessions/:sessionName/github/token", handlers.MintSessionGitHubToken)
projectGroup := api.Group("/projects/:projectName", handlers.ValidateProjectContext())
{
projectGroup.GET("/models", handlers.ListModelsForProject)
+ projectGroup.GET("/runner-types", handlers.GetRunnerTypes)
projectGroup.GET("/access", handlers.AccessCheck)
projectGroup.GET("/integration-status", handlers.GetProjectIntegrationStatus)
projectGroup.GET("/users/forks", handlers.ListUserForks)
diff --git a/components/backend/types/models.go b/components/backend/types/models.go
index f01caf124..3f38c527d 100644
--- a/components/backend/types/models.go
+++ b/components/backend/types/models.go
@@ -20,9 +20,10 @@ type ModelEntry struct {
// ModelManifest represents the top-level model manifest structure.
type ModelManifest struct {
- Version int `json:"version"`
- DefaultModel string `json:"defaultModel"`
- Models []ModelEntry `json:"models"`
+ Version int `json:"version"`
+ DefaultModel string `json:"defaultModel"`
+ ProviderDefaults map[string]string `json:"providerDefaults,omitempty"`
+ Models []ModelEntry `json:"models"`
}
// ListModelsResponse is the API response for the models endpoint.
diff --git a/components/frontend/src/app/admin/runtimes/page.tsx b/components/frontend/src/app/admin/runtimes/page.tsx
index 22dbc1576..86f1c5921 100644
--- a/components/frontend/src/app/admin/runtimes/page.tsx
+++ b/components/frontend/src/app/admin/runtimes/page.tsx
@@ -19,7 +19,7 @@ import { Skeleton } from "@/components/ui/skeleton";
import { Breadcrumbs } from "@/components/breadcrumbs";
import { PageHeader } from "@/components/page-header";
import { EmptyState } from "@/components/empty-state";
-import { useRunnerTypes } from "@/services/queries/use-runner-types";
+import { useRunnerTypesGlobal } from "@/services/queries/use-runner-types";
import type { RunnerType } from "@/services/api/runner-types";
function RuntimeStatusBadge() {
@@ -40,7 +40,7 @@ function RuntimeDetailPanel({ runtime }: { runtime: RunnerType }) {
Required keys:
- {(runtime.auth?.requiredSecretKeys ?? runtime.requiredSecretKeys ?? []).join(", ") || "None"}
+ {(runtime.auth?.requiredSecretKeys ?? []).join(", ") || "None"}
@@ -56,17 +56,10 @@ function RuntimeDetailPanel({ runtime }: { runtime: RunnerType }) {
-
Models
-
- {runtime.models.map((model) => (
-
- {model.label}
-
- ))}
-
-
- Default: {runtime.defaultModel}
-
+
Provider
+
+ {runtime.provider}
+
@@ -95,7 +88,7 @@ function LoadingSkeleton() {
}
export default function AdminRuntimesPage() {
- const { data: runtimes, isLoading, isError, error, refetch } = useRunnerTypes();
+ const { data: runtimes, isLoading, isError, error, refetch } = useRunnerTypesGlobal();
const [expandedId, setExpandedId] = useState
(null);
const toggleExpanded = (id: string) => {
@@ -174,7 +167,7 @@ export default function AdminRuntimesPage() {
Runtime
Description
- Models
+ Provider
Status
@@ -235,7 +228,7 @@ function RuntimeRow({
{runtime.description || "\u2014"}
- {runtime.models.length}
+ {runtime.provider}
diff --git a/components/frontend/src/app/api/projects/[name]/models/route.ts b/components/frontend/src/app/api/projects/[name]/models/route.ts
index 12edfa494..c1799a7cf 100644
--- a/components/frontend/src/app/api/projects/[name]/models/route.ts
+++ b/components/frontend/src/app/api/projects/[name]/models/route.ts
@@ -2,8 +2,9 @@ import { BACKEND_URL } from "@/lib/config";
import { buildForwardHeadersAsync } from "@/lib/auth";
/**
- * GET /api/projects/:projectName/models
- * Proxies to backend to list available models with workspace overrides
+ * GET /api/projects/:projectName/models?provider=...
+ * Proxies to backend to list available models with workspace overrides.
+ * Optional provider query param filters by model provider.
*/
export async function GET(
request: Request,
@@ -13,8 +14,12 @@ export async function GET(
const { name: projectName } = await params;
const headers = await buildForwardHeadersAsync(request);
+ const url = new URL(request.url);
+ const provider = url.searchParams.get("provider");
+ const backendParams = provider ? `?provider=${encodeURIComponent(provider)}` : "";
+
const response = await fetch(
- `${BACKEND_URL}/projects/${encodeURIComponent(projectName)}/models`,
+ `${BACKEND_URL}/projects/${encodeURIComponent(projectName)}/models${backendParams}`,
{ headers }
);
diff --git a/components/frontend/src/app/api/projects/[name]/runner-types/route.ts b/components/frontend/src/app/api/projects/[name]/runner-types/route.ts
new file mode 100644
index 000000000..85a719fed
--- /dev/null
+++ b/components/frontend/src/app/api/projects/[name]/runner-types/route.ts
@@ -0,0 +1,34 @@
+import { BACKEND_URL } from "@/lib/config";
+import { buildForwardHeadersAsync } from "@/lib/auth";
+
+/**
+ * GET /api/projects/:projectName/runner-types
+ * Proxies to backend to list available runner types with workspace overrides.
+ */
+export async function GET(
+ request: Request,
+ { params }: { params: Promise<{ name: string }> }
+) {
+ try {
+ const { name: projectName } = await params;
+ const headers = await buildForwardHeadersAsync(request);
+
+ const response = await fetch(
+ `${BACKEND_URL}/projects/${encodeURIComponent(projectName)}/runner-types`,
+ { headers }
+ );
+
+ const data = await response.text();
+
+ return new Response(data, {
+ status: response.status,
+ headers: { "Content-Type": "application/json" },
+ });
+ } catch (error) {
+ console.error("Failed to fetch runner types:", error);
+ return Response.json(
+ { error: "Failed to fetch runner types" },
+ { status: 500 }
+ );
+ }
+}
diff --git a/components/frontend/src/app/projects/[name]/sessions/[sessionName]/page.tsx b/components/frontend/src/app/projects/[name]/sessions/[sessionName]/page.tsx
index 17bf5f1b4..fa8b91594 100644
--- a/components/frontend/src/app/projects/[name]/sessions/[sessionName]/page.tsx
+++ b/components/frontend/src/app/projects/[name]/sessions/[sessionName]/page.tsx
@@ -234,7 +234,7 @@ export default function ProjectSessionDetailPage({
// Fetch runner capabilities and derive agent display name
const { data: capabilities } = useCapabilities(projectName, sessionName, phase === "Running");
- const { data: runnerTypes } = useRunnerTypes();
+ const { data: runnerTypes } = useRunnerTypes(projectName);
const agentName = useMemo(() => {
if (capabilities?.framework && runnerTypes) {
const matched = runnerTypes.find((rt) => rt.id === capabilities.framework);
diff --git a/components/frontend/src/components/create-session-dialog.tsx b/components/frontend/src/components/create-session-dialog.tsx
index 92329c0f5..a45930fdb 100644
--- a/components/frontend/src/components/create-session-dialog.tsx
+++ b/components/frontend/src/components/create-session-dialog.tsx
@@ -42,14 +42,8 @@ import { useIntegrationsStatus } from "@/services/queries/use-integrations";
import { useModels } from "@/services/queries/use-models";
import { errorToast } from "@/hooks/use-toast";
-// Keep in sync with components/manifests/base/models.json (available: true entries).
-const fallbackModels = [
- { value: "claude-sonnet-4-5", label: "Claude Sonnet 4.5" },
- { value: "claude-sonnet-4-6", label: "Claude Sonnet 4.6" },
- { value: "claude-opus-4-6", label: "Claude Opus 4.6" },
- { value: "claude-opus-4-5", label: "Claude Opus 4.5" },
- { value: "claude-haiku-4-5", label: "Claude Haiku 4.5" },
-];
+// Static default used for form initialization before the API responds.
+const DEFAULT_MODEL = "claude-sonnet-4-5";
const formSchema = z.object({
displayName: z.string().max(50).optional(),
@@ -76,16 +70,9 @@ export function CreateSessionDialog({
const [open, setOpen] = useState(false);
const router = useRouter();
const createSessionMutation = useCreateSession();
- const { data: runnerTypes, isLoading: runnerTypesLoading, isError: runnerTypesError, refetch: refetchRunnerTypes } = useRunnerTypes();
-
- const { data: modelsData, isLoading: modelsLoading } = useModels(projectName, open);
+ const { data: runnerTypes, isLoading: runnerTypesLoading, isError: runnerTypesError, refetch: refetchRunnerTypes } = useRunnerTypes(projectName);
const { data: integrationsStatus } = useIntegrationsStatus();
- const models = modelsData
- ? modelsData.models.map((m) => ({ value: m.id, label: m.label }))
- : fallbackModels;
- const defaultModel = modelsData?.defaultModel ?? "claude-sonnet-4-5";
-
const githubConfigured = integrationsStatus?.github?.active != null;
const gitlabConfigured = integrationsStatus?.gitlab?.connected ?? false;
const atlassianConfigured = integrationsStatus?.jira?.connected ?? false;
@@ -96,35 +83,44 @@ export function CreateSessionDialog({
defaultValues: {
displayName: "",
runnerType: DEFAULT_RUNNER_TYPE_ID,
- model: defaultModel,
+ model: DEFAULT_MODEL,
temperature: 0.7,
maxTokens: 4000,
timeout: 300,
},
});
- useEffect(() => {
- if (modelsData?.defaultModel && !form.formState.dirtyFields.model) {
- form.setValue("model", modelsData.defaultModel, { shouldDirty: false });
- }
- }, [modelsData?.defaultModel, form]);
-
const selectedRunnerType = form.watch("runnerType");
- // Derive the available models from the selected runner type
const selectedRunner = useMemo(
() => runnerTypes?.find((rt) => rt.id === selectedRunnerType),
[runnerTypes, selectedRunnerType]
);
- const availableModels = selectedRunner?.models ?? models;
+
+ // Fetch models filtered by the selected runner's provider.
+ // models.json is the single source of truth — no hardcoded fallback lists.
+ // Wait for runner types to load so we know the provider before fetching.
+ const { data: modelsData, isLoading: modelsLoading, isError: modelsError } = useModels(
+ projectName, open && !runnerTypesLoading && !runnerTypesError, selectedRunner?.provider
+ );
+
+ const models = modelsData
+ ? modelsData.models.map((m) => ({ value: m.id, label: m.label }))
+ : [];
+
+ // Update form model when API response arrives or provider changes
+ useEffect(() => {
+ if (modelsData?.defaultModel && !form.formState.dirtyFields.model) {
+ form.setValue("model", modelsData.defaultModel, { shouldDirty: false });
+ }
+ }, [modelsData?.defaultModel, form]);
const handleRunnerTypeChange = (value: string, onChange: (v: string) => void) => {
onChange(value);
- const runner = runnerTypes?.find((rt) => rt.id === value);
- if (runner) {
- // Reset model to the runner's default
- form.setValue("model", runner.defaultModel);
- }
+ // Model list will refetch via useModels when provider changes.
+ // resetField clears both value AND dirty state so the useEffect
+ // above will set the new provider's default model.
+ form.resetField("model", { defaultValue: "" });
};
const onSubmit = async (values: FormValues) => {
@@ -276,11 +272,17 @@ export function CreateSessionDialog({
- {availableModels.map((m) => (
-
- {m.label}
-
- ))}
+ {models.length === 0 && !modelsLoading ? (
+
+ No models available for this runner
+
+ ) : (
+ models.map((m) => (
+
+ {m.label}
+
+ ))
+ )}
@@ -433,7 +435,7 @@ export function CreateSessionDialog({
>
Cancel
-