Merge branch 'main' into refactor-inbound-context-routing-session

# Conflicts:
#	pkg/agent/eventbus_test.go
#	pkg/agent/loop.go
#	pkg/bus/bus.go
#	pkg/bus/types.go
#	pkg/channels/pico/pico.go
#	pkg/channels/telegram/telegram.go
#	pkg/config/config.go
#	web/backend/api/session.go
#	web/backend/api/session_test.go

pkg/agent/loop_test.go

@@ -2132,6 +2132,162 @@ func TestProcessMessage_ModelRoutingUsesLightProvider(t *testing.T) {
 	}
 }
 
+// TestProcessMessage_FallbackUsesPerCandidateProvider is the loop-level test for
+// bug #2140. It verifies that when the primary model returns a rate-limit error
+// the fallback closure routes the retry to the fallback model's own provider
+// (its own api_base), not back to the primary provider's endpoint.
+func TestProcessMessage_FallbackUsesPerCandidateProvider(t *testing.T) {
+	workspace := t.TempDir()
+
+	primaryCalls := 0
+	primaryServer := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
+		primaryCalls++
+		// Return 429 so FallbackChain classifies this as retriable and moves on.
+		w.Header().Set("Content-Type", "application/json")
+		w.WriteHeader(http.StatusTooManyRequests)
+		_ = json.NewEncoder(w).Encode(map[string]any{
+			"error": map[string]any{
+				"message": "rate limit exceeded",
+				"type":    "rate_limit_error",
+			},
+		})
+	}))
+	defer primaryServer.Close()
+
+	fallbackCalls := 0
+	fallbackServer := newStrictChatCompletionTestServer(
+		t, "fallback", "gemma-3-27b-it", "fallback reply", &fallbackCalls,
+	)
+	defer fallbackServer.Close()
+
+	cfg := &config.Config{
+		Agents: config.AgentsConfig{
+			Defaults: config.AgentDefaults{
+				Workspace:         workspace,
+				ModelName:         "mistral-primary",
+				ModelFallbacks:    []string{"gemma-fallback"},
+				MaxTokens:         4096,
+				MaxToolIterations: 3,
+			},
+		},
+		ModelList: []*config.ModelConfig{
+			{
+				ModelName: "mistral-primary",
+				Model:     "openrouter/mistralai/mistral-small-3.1",
+				APIBase:   primaryServer.URL,
+				APIKeys:   config.SimpleSecureStrings("primary-key"),
+				Workspace: workspace,
+			},
+			{
+				ModelName: "gemma-fallback",
+				Model:     "gemini/gemma-3-27b-it",
+				APIBase:   fallbackServer.URL,
+				APIKeys:   config.SimpleSecureStrings("fallback-key"),
+				Workspace: workspace,
+			},
+		},
+	}
+
+	provider, _, err := providers.CreateProvider(cfg)
+	if err != nil {
+		t.Fatalf("CreateProvider() error = %v", err)
+	}
+	msgBus := bus.NewMessageBus()
+	al := NewAgentLoop(cfg, msgBus, provider)
+	helper := testHelper{al: al}
+
+	resp := helper.executeAndGetResponse(t, context.Background(), bus.InboundMessage{
+		Channel:  "telegram",
+		SenderID: "user1",
+		ChatID:   "chat1",
+		Content:  "hi",
+	})
+
+	if resp != "fallback reply" {
+		t.Fatalf("response = %q, want %q (fallback provider)", resp, "fallback reply")
+	}
+	if primaryCalls == 0 {
+		t.Fatal("primary server was never called; expected at least one attempt")
+	}
+	if fallbackCalls != 1 {
+		t.Fatalf("fallback server calls = %d, want 1", fallbackCalls)
+	}
+}
+
+// TestProcessMessage_FallbackUsesActiveProviderWhenCandidateNotRegistered verifies
+// that when a candidate has no model_list entry it is absent from CandidateProviders
+// and the fallback closure falls back to activeProvider instead of panicking.
+func TestProcessMessage_FallbackUsesActiveProviderWhenCandidateNotRegistered(t *testing.T) {
+	workspace := t.TempDir()
+
+	// Primary server: returns 429 on first call, succeeds on second.
+	// Both the primary and the unregistered fallback share this server
+	// (same api_base) so activeProvider routes both calls here.
+	callCount := 0
+	primaryServer := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
+		callCount++
+		w.Header().Set("Content-Type", "application/json")
+		if callCount == 1 {
+			w.WriteHeader(http.StatusTooManyRequests)
+			_ = json.NewEncoder(w).Encode(map[string]any{
+				"error": map[string]any{"message": "rate limit", "type": "rate_limit_error"},
+			})
+			return
+		}
+		// Second call (fallback via activeProvider) succeeds.
+		_ = json.NewEncoder(w).Encode(map[string]any{
+			"choices": []map[string]any{
+				{"message": map[string]any{"content": "active provider reply"}, "finish_reason": "stop"},
+			},
+		})
+	}))
+	defer primaryServer.Close()
+
+	cfg := &config.Config{
+		Agents: config.AgentsConfig{
+			Defaults: config.AgentDefaults{
+				Workspace:         workspace,
+				ModelName:         "primary-model",
+				MaxTokens:         4096,
+				MaxToolIterations: 3,
+				// No model_list entry for this alias — absent from CandidateProviders.
+				ModelFallbacks: []string{"openrouter/fallback-model"},
+			},
+		},
+		ModelList: []*config.ModelConfig{
+			{
+				ModelName: "primary-model",
+				Model:     "openrouter/primary-model",
+				APIBase:   primaryServer.URL,
+				APIKeys:   config.SimpleSecureStrings("primary-key"),
+				Workspace: workspace,
+			},
+		},
+	}
+
+	provider, _, err := providers.CreateProvider(cfg)
+	if err != nil {
+		t.Fatalf("CreateProvider() error = %v", err)
+	}
+	msgBus := bus.NewMessageBus()
+	al := NewAgentLoop(cfg, msgBus, provider)
+
+	helper := testHelper{al: al}
+	resp := helper.executeAndGetResponse(t, context.Background(), bus.InboundMessage{
+		Channel:  "telegram",
+		SenderID: "user1",
+		ChatID:   "chat1",
+		Content:  "hi",
+	})
+
+	if resp != "active provider reply" {
+		t.Fatalf("response = %q, want %q", resp, "active provider reply")
+	}
+	if callCount < 2 {
+		t.Fatalf("primary server calls = %d, want >= 2 (one 429 + one success via activeProvider)", callCount)
+	}
+}
+
 // TestToolResult_SilentToolDoesNotSendUserMessage verifies silent tools don't trigger outbound
 func TestToolResult_SilentToolDoesNotSendUserMessage(t *testing.T) {
 	tmpDir, err := os.MkdirTemp("", "agent-test-*")