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 full‑featured chat UI with a “Tape” system that records every request, tool call and result so sessions can be paused, resumed, inspected, or replay‑backed later. The client can talk to any LLM that follows the OpenAI/Gemini/Anthropic API shape (cloud services like OpenAI, Gemini, Grok, etc.) or to locally hosted models via Ollama.

Why it matters – Most AI chat tools are thin web wrappers that lose context once you close the window, store data in the cloud, or require you to juggle many separate apps for tools, plugins, and remote bots. DeepChat keeps everything on your machine, gives you a single place to manage:

• Agents & Skills – reusable “Skills” (task‑specific instruction packs) that can be installed from folders, ZIPs or URLs and enabled per conversation.
• ACP integration – plug in any Agent Client Protocol‑compatible agent (coding assistants, task bots) and treat it like a regular model.
• MCP support – full Model Context Protocol handling of resources, prompts, tools, and streaming transports, with a built‑in Node.js toolchain for custom services.
• Remote control – bind a desktop session to Telegram, Discord, Feishu/Lark, QQBot or WeChat iLink so you can issue commands (/start, /stop, /model, …) from a messenger while the heavy work stays on your PC.
• Tape & Trace – a structured, recoverable log of the whole interaction, visible in a “Trace” view that shows token usage, provider metadata and the exact sequence of tool calls.

Key features at a glance

Category	Highlights
Local‑first UI	Multi‑window, multi‑tab, markdown + code‑block rendering (CodeMirror), image/Mermaid support, artifact preview, retry/fork conversation.
Model management	One place to configure dozens of cloud providers (OpenAI, Gemini, Anthropic, DeepSeek, etc.) and local Ollama models; download, start, stop models without a terminal.
Skills system	Install/import/export Skills; built‑in packs for code review, document handling, PDF/PowerPoint, frontend design, data‑sheet work, generative art, etc.
ACP agents	Load any ACP‑compatible agent as a selectable “model”; UI shows structured plans, tool calls and terminal output.
Remote bots	Telegram, Discord, Feishu/Lark, QQBot, WeChat iLink control – bind sessions, switch models, answer pending prompts, open the desktop UI remotely.
Search extensions	Integrated BoSearch / Brave Search APIs; can simulate browsing of Google, Bing, Baidu, Sogou, or any custom engine for LLM‑driven web queries.
MCP tooling	Resources / Prompts / Tools, streamable HTTP, SSE, stdio, in‑memory services, one‑click DeepLink installation of custom services.
Privacy & security	Local encrypted storage, optional network proxy, screen‑projection hiding, code‑obfuscation hooks.
Business‑ready	Apache‑2.0 license, clear separation of UI and backend, easy to embed or extend for commercial products.

Typical use cases

• Software development assistant – load a “Code Review” Skill, attach the ACP coding agent, and let the model edit, lint and commit code while you keep a full trace of what was changed.
• Document & data processing – enable a PDF‑Skill to let the agent read, summarize or extract tables from large documents, with results rendered as markdown tables or XLSX artifacts.
• Team‑wide bots – expose a DeepChat session to a Discord channel; teammates can ask the agent to run scripts, fetch data, or generate reports without leaving chat.
• Research & web‑augmented QA – turn on the built‑in search extension; the model can decide when to query Google/Bing, retrieve snippets and cite them directly in the conversation.
• Local‑only confidential work – run Ollama models on‑premise, keep all prompts and outputs on the machine, and still enjoy the same UI features as cloud‑based agents.

Getting started

1. Download – Grab the latest installer from the Releases page (Windows .exe, macOS .dmg, Linux .AppImage or .deb).
2. Configure a model – Open Settings → Model Providers, add your API key for OpenAI, Gemini, etc., or enable Ollama and select a locally installed model.
3. Add Skills (optional) – In Settings → Skills click Install and point to a folder/ZIP/URL containing a Skill pack.
4. Start a conversation – Click New Chat, pick a model or an ACP agent, and begin typing. Use the toolbar to attach files, trigger a search, or switch Skills on the fly.
5. Remote control – Enable a messenger channel under Settings → Remote, follow the on‑screen pairing steps, then issue /help in that messenger to see available commands.

Development & contribution

• The codebase is an Electron + React front‑end with a Node.js backend that implements the Tape, MCP and ACP layers.
• Run npm install then npm run dev to launch the app in development mode.
• Build installers with npm run build (cross‑platform scripts are provided).
• Contributions are welcomed via pull requests; see the CONTRIBUTING.md for style guidelines and how to add new Skills or ACP agents.

License – Apache License 2.0 (permissive for both personal and commercial use).

---

DeepChat aims to be the “Swiss‑army‑knife” desktop client for modern LLM agents, giving power users a recoverable, extensible and privacy‑first environment without the friction of juggling multiple tools.