feat(llm): quota-triggered model fallback (GH#8998)

autobot-backend/api/anthropic_compat.py

@@ -321,6 +321,8 @@ async def messages(
         )
 
     # Non-streaming path — use fallback coordinator for quota-triggered model switch
+    from llm_shared.model_fallback_coordinator import get_fallback_coordinator
+
     llm_response = await get_fallback_coordinator().execute_with_fallback(llm_request, registry)
     if llm_response.error:
         raise HTTPException(status_code=502, detail=llm_response.error)

autobot-backend/api/chat.py

@@ -36,6 +36,8 @@
     ChatPreferences,
     ChatSaveData,
     ChatStatsData,
+    ConversationSummarizeData,
+    ConversationSummarizeRequest,
     DetectLanguageData,
     DetectLanguageRequest,
     EnhancedChatCapabilitiesData,
@@ -2187,3 +2189,88 @@ async def detect_language(
     )
     result = await agent.handle_detect_language(req)
     return JSONResponse(content=result, media_type="application/json; charset=utf-8")  # codeql[py/stack-trace-exposure]
+
+
+@router.post("/chat/summarize", response_model=DataResponse[Dict[str, Any]])
+@with_error_handling(
+    category=ErrorCategory.SERVER_ERROR,
+    operation="summarize_conversation",
+    error_code_prefix="CHAT_SUMMARIZE",
+)
+async def summarize_conversation(
+    body: "ConversationSummarizeRequest",
+    request: Request,
+    current_user: dict = Depends(get_current_user),
+):
+    """
+    Summarize a range of conversation messages for context overflow protection (GH#9043).
+
+    When a conversation approaches the model's context window limit, this endpoint
+    generates a compact summary of the earliest messages to enable a rolling window
+    approach without losing essential context.
+    """
+    from api.schemas_chat import ConversationSummarizeData, ConversationSummarizeRequest
+
+    request_id = generate_request_id()
+    logger.info("[%s] Summarizing %d messages from session %s", request_id, len(body.message_ids), body.session_id)
+
+    # Validate session ownership
+    await validate_chat_ownership(body.session_id, request)
+
+    # Get chat history manager and retrieve messages
+    chat_history = get_chat_history_manager()
+    all_messages = await chat_history.get_messages(body.session_id)
+
+    # Filter to requested message IDs
+    messages_to_summarize = [msg for msg in all_messages if msg.get("message_id") in body.message_ids]
+
+    if not messages_to_summarize:
+        raise HTTPException(status_code=404, detail="No messages found with the provided IDs")
+
+    # Build prompt for summarization
+    conversation_text = "\n\n".join(
+        [f"{msg.get('role', 'unknown').upper()}: {msg.get('content', '')}" for msg in messages_to_summarize]
+    )
+
+    summarization_prompt = f"""Summarize the following conversation segment concisely. Preserve key context, decisions, and technical details that would be needed to continue the conversation naturally.
+
+Target length: approximately {body.target_length or 500} tokens.
+
+Conversation:
+{conversation_text}
+
+Summary:"""
+
+    # Get LLM service and generate summary
+    llm_service = get_llm_service(request)
+
+    try:
+        summary_response = await llm_service.generate_response(
+            messages=[{"role": "user", "content": summarization_prompt}],
+            model="gpt-4o-mini",  # Fast, cheap model for summarization
+            max_tokens=body.target_length or 500,
+            temperature=0.3,  # Lower temperature for focused summarization
+        )
+
+        summary_text = summary_response.get("content", "")
+        summary_tokens = summary_response.get("usage", {}).get("total_tokens", 0)
+
+    except Exception as e:
+        logger.exception("[%s] Failed to generate summary", request_id)
+        raise HTTPException(status_code=500, detail=f"Summarization failed: {str(e)}")
+
+    result_data = ConversationSummarizeData(
+        summary=summary_text,
+        original_message_count=len(messages_to_summarize),
+        summary_token_count=summary_tokens,
+        timestamp=utc_timestamp(),
+    )
+
+    logger.info(
+        "[%s] Generated summary: %d messages → %d tokens",
+        request_id,
+        result_data.original_message_count,
+        result_data.summary_token_count or 0,
+    )
+
+    return DataResponse(data=result_data.model_dump())

autobot-backend/api/openai_compat.py

@@ -381,6 +381,8 @@ async def chat_completions(
         )
 
     # Non-streaming path — use fallback coordinator for quota-triggered model switch
+    from llm_shared.model_fallback_coordinator import get_fallback_coordinator
+
     llm_response = await get_fallback_coordinator().execute_with_fallback(llm_request, registry)
     if llm_response.error:
         raise HTTPException(status_code=502, detail=llm_response.error)

autobot-backend/api/schemas_agent.py

@@ -39,6 +39,7 @@ class AgentCommandExecuteResponse(BaseModel):
     output: str | None = None
     status: str | None = None
 
+
 class AgentHealthResponse(BaseModel):
     """Response for GET /health/enhanced."""
 
@@ -65,6 +66,7 @@ class MemoryStatisticsResponse(BaseModel):
     active_tasks: Any | None = None
     performance_insights: Any | None = None
 
+
 class MemoryTaskHistoryResponse(BaseModel):
     """Response for GET /tasks/history."""
 
@@ -123,6 +125,7 @@ class MemoryDocumentReferencesResponse(BaseModel):
     timestamp: str
     references: Any | None = None
 
+
 class MemoryEmbeddingCacheStatsResponse(BaseModel):
     """Response for GET /embeddings/cache-stats."""
 

autobot-backend/api/schemas_chat.py

@@ -669,3 +669,25 @@ class SessionFolderAssign(BaseModel):
     """Request body for PUT /chat/sessions/{session_id}/folder."""
 
     folder_id: str | None = Field(None, description="Folder ID to assign; None removes from folder")
+
+
+# ── Context Overflow Protection schemas (GH#9043) ────────────────────────────
+
+
+class ConversationSummarizeRequest(BaseModel):
+    """Request body for POST /chat/summarize."""
+
+    session_id: str = Field(..., description="Chat session ID")
+    message_ids: List[str] = Field(..., min_length=1, description="Message IDs to summarize")
+    target_length: int | None = Field(
+        500, ge=100, le=2000, description="Target summary length in tokens (default: 500)"
+    )
+
+
+class ConversationSummarizeData(BaseModel):
+    """data payload for POST /chat/summarize."""
+
+    summary: str = Field(..., description="Generated summary of the conversation segment")
+    original_message_count: int = Field(..., description="Number of messages summarized")
+    summary_token_count: int | None = Field(None, description="Token count of the summary")
+    timestamp: str = Field(..., description="ISO timestamp when summary was generated")

autobot-backend/api/schemas_code.py

@@ -225,6 +225,7 @@ class LogRecentResponse(BaseModel):
     source: str
     error: str | None = None
 
+
 class LogReadResponse(BaseModel):
     """Response for GET /logs/read/{filename}."""
 
@@ -282,6 +283,7 @@ class GitMCPOperationResponse(BaseModel):
     timestamp: str
     errors: str | None = None
 
+
 class GitMCPInfoResponse(BaseModel):
     """Response for GET /git/mcp/info."""
 
@@ -351,6 +353,7 @@ class SkillsGapResponse(BaseModel):
     name: str | None = None
     tools_found: List[str] | None = None
 
+
 class SkillsDraftListItem(BaseModel):
     """Single draft entry in the list response."""
 
@@ -361,6 +364,7 @@ class SkillsDraftListItem(BaseModel):
     created_at: Any | None = None
     trust_level: Any | None = None
 
+
 class SkillsDraftTestResponse(BaseModel):
     """Response for POST /skills-governance/drafts/{skill_id}/test."""
 
@@ -536,18 +540,21 @@ class AccessControlMetricsResponse(BaseModel):
     success: bool
     data: Dict[str, Any] | None = None
 
+
 class AccessControlEndpointMetricsResponse(BaseModel):
     """Response for GET /access-control/endpoint/{endpoint:path}."""
 
     success: bool
     data: Any | None = None
 
+
 class AccessControlUserMetricsResponse(BaseModel):
     """Response for GET /access-control/user/{username}."""
 
     success: bool
     data: Any | None = None
 
+
 class AccessControlCleanupResponse(SuccessMessageResponse):
     """Response for POST /access-control/cleanup."""
 

autobot-backend/api/schemas_common.py

@@ -41,6 +41,7 @@ class SuccessDataResponse(BaseModel):
     message: str
     data: Dict[str, Any] | None = None
 
+
 class SuccessResponse(BaseModel):
     """Minimal success/failure envelope used by several endpoints."""
 

autobot-backend/api/schemas_terminal.py

@@ -25,6 +25,7 @@ class TerminalSessionCreateResponse(BaseModel):
     created_at: str
     ssh_keys: Dict[str, Any] | None = None
 
+
 class TerminalSessionItem(BaseModel):
     """Single session entry within a list response."""
 
@@ -72,6 +73,7 @@ class SSHKeyItem(BaseModel):
     has_passphrase: bool | None = None
     key_path: str | None = None
 
+
 class SSHKeyListResponse(BaseModel):
     """Response for GET /sessions/{session_id}/ssh-keys."""
 
@@ -143,6 +145,7 @@ class TerminalCommandHistoryResponse(BaseModel):
     total_commands: int | None = None
     message: str | None = None
 
+
 class TerminalAuditLogResponse(BaseModel):
     """Response for GET /audit/{session_id}."""
 
@@ -190,6 +193,7 @@ class TerminalHealthResponse(BaseModel):
     metrics: TerminalHealthMetrics | None = None
     error: str | None = None
 
+
 class TerminalStatusFeatures(BaseModel):
     pty_support: bool
     websocket_support: bool
@@ -203,6 +207,7 @@ class TerminalStatusSessionInfo(BaseModel):
     active_sessions: int
     max_concurrent_sessions: int | None = None
 
+
 class TerminalSystemStatusResponse(BaseModel):
     """Response for GET /status."""
 
@@ -212,6 +217,7 @@ class TerminalSystemStatusResponse(BaseModel):
     session_info: TerminalStatusSessionInfo | None = None
     error: str | None = None
 
+
 class TerminalCapabilitiesResponse(BaseModel):
     """Response for GET /capabilities."""
 
@@ -244,6 +250,7 @@ class TerminalImplementationItem(BaseModel):
     approval_workflow: bool
     service_layer: str | None = None
 
+
 class TerminalFeaturesResponse(BaseModel):
     """Response for GET /features."""
 
@@ -315,6 +322,7 @@ class AgentTerminalSessionCreateResponse(BaseModel):
     created_at: float
     pty_session_id: str | None = None
 
+
 class AgentTerminalSessionItem(BaseModel):
     """Single session entry within list response."""
 
@@ -329,6 +337,7 @@ class AgentTerminalSessionItem(BaseModel):
     command_count: int
     pty_session_id: str | None = None
 
+
 class AgentTerminalSessionListResponse(BaseModel):
     """Response for GET /agent-terminal/sessions."""
 
@@ -376,6 +385,7 @@ class AgentTerminalCommandStateResponse(BaseModel):
     approved_by_user_id: str | None = None
     approval_comment: str | None = None
 
+
 class AgentTerminalInfoResponse(BaseModel):
     """Response for GET /agent-terminal/."""
 
@@ -416,6 +426,7 @@ class AgentTerminalHostSelectionGetResponse(BaseModel):
     created_at: str
     updated_at: str | None = None
 
+
 class AgentTerminalHostSelectionSubmitResponse(BaseModel):
     """Response for POST /agent-terminal/host-selection/{request_id}/select."""
 
@@ -425,6 +436,7 @@ class AgentTerminalHostSelectionSubmitResponse(BaseModel):
     selected_host_name: str | None = None
     connection_info: Dict[str, Any] | None = None
 
+
 class AgentTerminalHostSelectionCancelResponse(BaseModel):
     """Response for POST /agent-terminal/host-selection/{request_id}/cancel."""