ThinkInAIXYZ/deepchat

DeepChat – Open‑Source Local‑First AI Agent Desktop Client

What it is – DeepChat is a cross‑platform desktop application (Windows, macOS, Linux) that lets you run LLM‑powered “agents” locally. It bundles a UI, a session‑recording system (called Tape), a plug‑in mechanism (Skills), and integrations for remote control via popular messaging platforms. It works with any provider that speaks the OpenAI/Gemini/Anthropic API format, as well as locally hosted models through Ollama or LM Studio.

Key ideas

• Local‑first – the core app runs on your machine; only the model API calls go over the network (or stay completely offline with Ollama).
• Tape.systems philosophy – every conversation is recorded as a structured “tape” that can be inspected, resumed, or replayed, making long‑running agent workflows reproducible.
• Skills – reusable bundles of instructions, reference files, and optional scripts that can be installed from a folder, ZIP, or URL and then enabled per‑chat.
• ACP (Agent Client Protocol) support – external agent runtimes that implement ACP appear as selectable “models” inside the UI, giving you a native UI for coding or task‑oriented agents.
• Remote control – a session can be bound to a Telegram, Feishu/Lark, QQBot, Discord, or WeChat iLink endpoint, allowing you to issue commands (e.g., /new, /stop, /model) from those messengers.

Main features

• Multi‑model management (cloud APIs such as OpenAI, Gemini, Anthropic, DeepSeek, etc., and local Ollama models).
• Rich markdown rendering with CodeMirror code blocks, Mermaid diagrams, image and multi‑modal output (including GPT‑4o‑style text‑to‑image).
• Session “tape” and “trace” views that show request sequences, token budgets, and provider metadata.
• Installable Skills for domains like code review, document handling (PDF/DOCX), front‑end design, spreadsheet generation, etc.
• Built‑in search extensions (BoSearch, Brave, Google, Bing, Baidu, Sogou) that let the LLM invoke web searches as a human would.
• Strong MCP (Model Context Protocol) support: resources, prompts, tools, streamable HTTP/SSE, stdio services, and a Node.js toolchain for custom services.
• Security‑first design – optional encryption of chat/config data, code obfuscation, network‑proxy support, and screen‑projection hiding.
• Business‑friendly Apache‑2.0 license, with hooks for enterprise‑grade encrypted/obfuscated deployments.

Typical use cases

• Developer assistant – run an ACP‑compatible coding agent, browse file system, execute code, and keep the session state across restarts.
• Research & writing – enable a “document‑review” Skill, attach PDFs or DOCX files, and let the agent cite and summarize them.
• Team bot – expose a DeepChat session to a Discord or Telegram channel so teammates can ask the agent questions without leaving their chat app.
• Local‑only privacy‑sensitive work – use Ollama models entirely offline while still benefitting from the UI, Skills, and tape‑based session recovery.

Getting started

1. Download – Grab the latest installer from the GitHub Releases page (Windows .exe, macOS .dmg, Linux .AppImage).
2. Configure a model – Open Settings → Model Providers, add an API key for OpenAI, Gemini, etc., or enable the built‑in Ollama manager to pull a local model.
3. Start a conversation – Choose a model (or an ACP agent) from the selector, type a prompt, and watch the tape/trace panels fill in as the agent works.
4. Add Skills – In Settings → Skills, install a Skill package (folder/ZIP/URL) and toggle it on for the current chat.
5. Remote control (optional) – Enable a messenger channel under Settings → Remote, pair it with a session, and issue /help in the messenger to see available commands.

Development

• The repo uses Electron for the desktop shell and React for the renderer.
• Dependencies can be installed with npm install (or yarn).
• Run a hot‑reload dev build via npm run dev.
• Build a production package with npm run build.

Community & contribution

• Open issues for bugs or feature ideas.
• Pull requests are welcomed; the project follows the Apache‑2.0 license, so commercial use is permitted without copyleft restrictions.

License – Apache License 2.0.

---

All details above are taken directly from the repository’s README; no additional features have been inferred.