ProvidersAgentsFeaturesDocsDownload GitHubDownload
Open source Apache 2.0

The control plane for AI coding agents.

Run agents, models, MCP tools, and skills in one workspace. Bring your own keys, keep your workflow, own the stack.

Download
brew install --cask argos
Argos desktop app showing a multi-model chat session with tool calls

Works with the providers you already use

  • DeepSeek
  • Google Gemini
  • Anthropic
  • xAI Grok
  • GitHub Models
  • Ollama
  • OpenRouter
  • Mistral
  • Hugging Face
  • Vertex AI
  • DashScope
  • Llama

Plus OpenAI, AWS Bedrock, Azure, and any OpenAI, Gemini, or Anthropic-compatible endpoint. That is 40+ providers in one interface.

Agent-first

Made for agents to plug in.

Argos speaks the protocols your agents already use. Three surfaces, no vendor lock-in, all running on your machine.

  • Connect ACP agents

    Any ACP-compatible runtime becomes a first-class entry in the model selector, with a native workspace UI. Enable a built-in agent or add your own command.

    Settings → ACP Agents

  • Attach MCP tools

    Full Resources, Prompts, and Tools coverage with visual debugging when a server misbehaves. A bundled Bun runtime executes servers without extra setup.

    StreamableHTTPSSEstdio

  • Install Skills

    Standard Agent Skills spec. Install from folders, ZIPs, or URLs, and import or export packs with Claude Code, Codex, Cursor, and GitHub Copilot.

    Settings → Skills

Driving Argos from your own tooling? The daemon exposes POST /api/v1/route and a WebSocket event stream at /api/v1/events. See the daemon handbook.

One window for the whole workflow

No browser tab sprawl, no context switching between five chat windows. The pieces sit next to each other.

Multi-LLM chat

DeepSeek, Gemini, Anthropic, Grok, Ollama, and any OpenAI-, Gemini-, or Anthropic-compatible endpoint. Switch models per conversation.

40+

Browser workspace

The daemon serves the same chat and workspace shell over HTTP. Pair a browser session and leave Electron closed.

Built-in search

BoSearch and Brave Search ship as MCP integrations; add any engine via a search-assistant model.

BoSearchBrave SearchCustom engine

Local and offline

Integrated Ollama: download, deploy, and run models fully offline.

Rich rendering

Markdown, syntax highlighting, Mermaid diagrams, and Artifacts in every conversation.

Parallel sessions

Multi-window, multi-tab parallelism with retry, regenerate, and conversation forks.

Step away. Stay in the loop.

A session does not stop when you leave your desk. Drive it from a paired browser or your messaging app.

TelegramDiscordFeishu / LarkQQBotWeChat iLink
Argos chat workspace in light mode showing rendered markdown, code, and a diagram
  • Pair over a one-time URL
  • Answer pending prompts
  • Switch models remotely
  • Stop a runaway generation
  • Check runtime status
  • Open the session on desktop

Get Argos

Free and open source. Pick your platform.

Windows.exe installer
macOSHomebrew or .dmg
Linux.AppImage / .deb

Run it headless

Desktop is all most users need. Install Argos Server only when another machine should run agents and host project files.

macOS / Linuxcurl -fsSL https://raw.githubusercontent.com/dvaJi/argos/main/distro/install/install.sh | sh
Windowsirm https://raw.githubusercontent.com/dvaJi/argos/main/distro/install/install.ps1 | iex
Homebrewbrew install dvaJi/tap/argos-daemon

Then run argos-daemon --web --pair to serve the web workspace. Deploying on a VPS? Read the daemon handbook first.