This is an early release preview. You may encounter bugs.
DeepChat logo
Unclaimed

App interface connectors

DeepChat

Local-first AI agent desktop client with MCP, Skills, ACP agents, and remote control from messaging apps

No votes yet

01 / About

What DeepChat is.

DeepChat is a local-first desktop client that puts chat models, tools, skills, and agent runtimes in one Electron application for Windows, macOS, and Linux. It talks to cloud APIs and to locally deployed Ollama models through the same interface, stores chat and configuration data locally, and supports network proxies and screen-projection hiding.

Sessions follow the Tape.systems approach: a session Tape records structured work history so context, tool calls, requests, and results stay recoverable and resumable, and trace previews expose the request sequence, provider and model metadata, Tape view manifests, included entries, and token budgets for long-running agent work.

Three extension surfaces sit alongside the chat window. Skills follow the Agent Skills specification and can be installed from folders, ZIP files, or URLs, enabled per conversation, and exchanged with other compatible tools; built-in skills cover generative art, code review, document collaboration, DOCX, PDF, PPTX, XLSX, frontend design, git commit messages, MCP building, and web artifacts. Agent Client Protocol (ACP) support runs external agent runtimes as first-class entries in the model selector, with a workspace view for plans, tool calls, and terminal output. Model Context Protocol (MCP) support covers resources, prompts, and tools over StreamableHTTP, SSE, and stdio transports, plus in-memory services for code execution, web retrieval, and file operations, and one-click installation through DeepLinks.

Remote control binds a running session to a messaging channel — Telegram, Feishu/Lark, QQBot, Discord, or WeChat iLink — so you can start sessions, switch models, answer pending permission prompts, stop a run, or open the session on the desktop with commands such as /new, /sessions, /use, /stop, /pending, and /status.

Features

  • Session Tape and trace: recoverable, inspectable work history with request sequences, model metadata, and token budgets
  • Skills: install from folders, ZIP files, or URLs; enable per conversation; import and export with other Agent Skills tools
  • ACP agents: run Agent Client Protocol runtimes as selectable models, with a structured plan and tool-output view
  • MCP support: resources, prompts, and tools over multiple transports, plus in-memory services and DeepLink installation
  • Remote control: drive a session from Telegram, Feishu/Lark, QQBot, Discord, or WeChat iLink
  • Provider breadth: DeepSeek, OpenAI, Anthropic, Gemini, Moonshot, Grok, and many aggregators, plus anything speaking the OpenAI, Gemini, or Anthropic API format
  • Local models: integrated Ollama management for downloading, deploying, and running models without the command line
  • Multi-window chat: parallel tabs and windows, message retry variants, and forkable conversations
  • Rich rendering: Markdown with CodeMirror-based code blocks, artifacts, images, and Mermaid diagrams
  • Search extension: built-in BoSearch and Brave Search integration, plus browsing-based access to engines without an API
  • Privacy controls: local data storage, network proxy support, screen-projection hiding, and reserved encryption interfaces

02 / Discussion CREDIBILITY-GATED

Discussion

Reading is open to everyone. Posting and voting need a verified identity or a GitHub grade of B or higher.

  • No discussions yet.

04 / Build

Build with DeepChat.

Browse the catalogue for frameworks, tools, and harnesses, each scored on real GitHub credibility.

Get DeepChat →

Browse the catalogue