You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
MCP (Model Context Protocol) full pipeline — main-process multi-connection manager supporting stdio, SSE, and streamable HTTP transports; IPC endpoints for lifecycle control; renderer-side tool bridge that injects MCP tools into the agent loop.
Settings → MCP panel — create, edit, and manage MCP server configs with transport-aware forms, live capability refresh, and connection status controls.
Chat composer MCP awareness — MCP servers appear in the Skills menu and as inline badges; users can toggle MCP capabilities per session before sending a message.
Gitee AI provider preset refresh — curated DeepSeek, Qwen, GLM, Kimi, MiniMax, ERNIE, and Hunyuan models with accurate token limits and thinking support flags.
System prompt language injection — agent system prompt now includes the user's selected language so AI responses match the UI locale.
Animated transitions component (animate-ui/transitions.tsx) for smoother UI state changes.
Confirm dialog component (ui/confirm-dialog.tsx) for destructive action confirmations.
Changed
README repositioned as the open-source Claude Cowork alternative; added provider comparison table, download links, and streamlined quick-start guide.
Locales (EN + ZH) gained MCP-related strings, ensuring consistent translations across chat and settings screens.
Settings page restructured with tabbed layout and expanded provider configuration.
AssistantMessage, ThinkingBlock, ToolCallCard, and FileChangeCard components refactored for cleaner rendering and better streaming display.
Layout simplified — Layout.tsx reduced by ~400 lines; panel logic extracted into dedicated components.
Chat store enhanced with improved session management and message persistence.
Fixed
Chat input warnings and badges now reliably reflect active providers, MCP servers, and context window usage before running agent tasks.