Add multi-provider web search support: Brave (default), Tavily,
DuckDuckGo, and SearXNG. Falls back to DuckDuckGo when provider
credentials are missing. Providers are dispatched via a map with
register_provider() for plugin extensibility.
- WebSearchConfig with env-var resolution and from_legacy() bridge
- Config migration for legacy flat keys (tavilyApiKey, searxngBaseUrl)
- SearXNG URL validation, explicit error for unknown providers
- ddgs package (replaces deprecated duckduckgo-search)
- 16 tests covering all providers, fallback, env resolution, edge cases
- docs/web-search.md with full config reference
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
- Add group_policy config: 'open' (default) or 'mention'
- 'open': Respond to all group messages (backward compatible)
- 'mention': Only respond when @mentioned in any group
- Auto-detect bot mentions by pattern matching:
* If open_id configured: match against mentions
* Otherwise: detect bot by empty user_id + ou_ open_id pattern
- Support @_all mentions
- Private chats unaffected (always respond)
- Clean implementation with minimal logging
docs: update Feishu README with group policy documentation
MCP SDK's anyio cancel scopes can leak CancelledError on timeout or
failure paths. Since CancelledError is a BaseException (not Exception),
it escapes both MCPToolWrapper.execute() and ToolRegistry.execute(),
crashing the agent loop.
Now catches CancelledError and returns a graceful error to the LLM,
while still re-raising genuine task cancellations from /stop.
Also catches general Exception for other MCP failures (connection
drops, invalid responses, etc.).
Related: #1055
Major changes:
- Replace message-count-based memory window with token-budget-based compression
- Add max_tokens_input, compression_start_ratio, compression_target_ratio config
- Implement _maybe_compress_history() that triggers based on prompt token usage
- Use _build_compressed_history_view() to provide compressed history to LLM
- Refactor MemoryStore.consolidate() -> consolidate_chunk() for chunk-based compression
- Remove last_consolidated from Session, use _compressed_until metadata instead
- Add background compression scheduling to avoid blocking message processing
Key improvements:
- Compression now based on actual token usage, not arbitrary message counts
- Better handling of long conversations with large context windows
- Non-destructive compression: old messages remain in session, but excluded from prompt
- Automatic compression when history exceeds configured token thresholds
