MCP 및 도구 호출

DeepChat - Open-Source Local-First AI Agent Desktop Client

thinkinaixyz/deepchat

DeepChat is an open-source, local-first AI agent desktop client built in TypeScript and Electron. It integrates cloud LLMs, local models, MCP services, installable Skills, ACP agents, and remote control for messaging apps, following the Tape.systems philosophy to keep agent sessions recoverable and inspectable.

★ 6.2K별점
⑂ 712포크 수
6미해결 이슈
TypeScript언어
Apache-2.0라이선스
Q@project.QualityScore편집 점수

프로젝트 스크린샷

DeepChat - Open-Source Local-First AI Agent Desktop Client 스크린샷 DeepChat - Open-Source Local-First AI Agent Desktop Client 스크린샷

개요

DeepChat unifies multi-model management, agent runtimes, and long-running sessions into a single desktop application. It supports a wide range of cloud LLM providers and local Ollama models. The application is designed around the Tape.systems philosophy, preserving structured session history, request context, tool calls, and token budgets for debugging and recovery. It features robust MCP support, ACP integration for external coding/task agents, and remote control capabilities via Telegram, Feishu/Lark, QQBot, Discord, and WeChat iLink.

주요 기능

  • Local-first AI agent desktop client
  • Tape.systems philosophy for recoverable and traceable sessions
  • Installable and portable Skills compatible with Claude Code, Codex, Cursor, and more
  • Native ACP (Agent Client Protocol) integration
  • Strong MCP support with Resources, Prompts, Tools, and multiple transports
  • Remote control via Telegram, Feishu/Lark, QQBot, Discord, and WeChat iLink
  • Unified multi-model management for cloud and local models
  • Multi-window and multi-tab parallel session architecture
  • Artifacts rendering, Mermaid diagrams, and multi-modal content support
  • Built-in search integration with BoSearch, Brave Search, and simulated web browsing for Google, Bing, Baidu
  • DeepLink support for one-click MCP installation and conversation initiation
  • Cross-platform support for Windows, macOS, and Linux

요구 사항, 설치 및 빠른 시작

End-users can install via GitHub Releases (.exe for Windows, .dmg for macOS, .AppImage/.deb for Linux), the official website, or Homebrew for macOS (brew install --cask deepchat). For development: run 'pnpm install' followed by 'pnpm run installRuntime'. If 'No module named distutils' error occurs, run 'pip install setuptools'.

사용 정보

After installation, launch the app, open settings, and configure Model Providers with API keys or local Ollama. Create a new conversation with the '+' button, select a model, and begin chatting. Skills can be managed via Settings -> Skills, ACP agents via Settings -> ACP Agents, and remote channels via Settings -> Remote.

모델 호환성 및 사용 사례

Compatible with any model provider using OpenAI, Gemini, or Anthropic API formats. Explicitly supported providers include DeepSeek, OpenAI, Moonshot/Kimi, Grok, Gemini, Anthropic, Ollama, Qiniu, New API, Zhipu, PPIO, MiniMax, Fireworks, AIHubMix, Doubao, DashScope, Groq, JieKou.AI, ZenMux, GitHub Models, LM Studio, Hunyuan, 302.AI, Together, Poe, Vercel AI Gateway, OpenRouter, Azure OpenAI, TokenFlux, BurnCloud, OpenAI Responses, CherryIn, ModelScope, AWS Bedrock, Voice.ai, Vertex AI, GitHub Copilot, Xiaomi, o3.fan, Novita AI, Astraflow, SiliconFlow, and OrcaRouter.

라이선스 및 위험 참고 사항

Licensed under the Apache License 2.0. Suitable for both commercial and personal use. Enterprise integration is supported with minimal configuration code changes to utilize reserved encrypted obfuscation security capabilities.

Editorial verification 2026-08-09: repository URL, owner, description, license and repository statistics were reviewed. License metadata: Apache-2.0. README was fetched for the channel draft; re-check repository dependencies, releases and model terms before production use.

릴리스 및 유지 관리

The repository was created on 2025-02-14. It currently has 6205 stars, 712 forks, and 6 open issues. The default branch is 'dev'. The last push was recorded on 2026-08-09.

MemPalace

mempalace/mempalace

★ 58KPython

QwenPaw

agentscope-ai/qwenpaw

★ 33.9KPython

Composio

composiohq/composio

★ 29.5KTypeScript