← Back to Discover

SwarmClaw

Open-source runtime you host yourself for running AI agents and agent teams, with schedules, delegation, durable memory, MCP tools and chat connectors such as Discord, Slack and Telegram.

Agent frameworksAutonomous agents
Visit website ↗

What it does

SwarmClaw is an open-source agent runtime that you host yourself. It can run as a desktop app for macOS, Windows and Linux, as a command-line install through npm, or in Docker. Starting it brings up a local server on port 3456 with a dashboard for creating agents, chatting with them and managing tasks. Agents can run on schedules and heartbeat loops, keep memory between conversations, and hand work to other agents. Delegation can also target coding tools such as Claude Code, Codex CLI, OpenCode CLI and Gemini CLI, as well as SwarmClaw's own subagents. An org-chart view and structured sessions help organise teams of agents. The site lists more than 20 built-in model providers, including Anthropic, OpenAI, Google Gemini, OpenRouter, DeepSeek, Ollama and LM Studio, plus compatible custom endpoints. MCP servers can be connected, and connectors reach Discord, Slack, Telegram, WhatsApp, Teams, Matrix and email. It is released under the MIT licence. Setup needs Node.js 22.6 or newer and npm 10 or newer, and Docker Desktop is recommended if you want sandboxed browser execution.

Capabilities

  • Self-hosted agent runtime available as a desktop app, an npm command-line install or a Docker container
  • Schedules, heartbeat loops and background jobs for agents that keep working without prompting
  • Delegation between agents, including to Claude Code, Codex CLI, OpenCode CLI and Gemini CLI
  • Durable memory and runtime skills that carry context across conversations
  • MCP server connections and chat connectors for Discord, Slack, Telegram, WhatsApp, Teams and Matrix
  • More than 20 built-in model providers plus compatible custom endpoints

When to use it

  • Running your own always-on agents on hardware you control instead of a hosted service
  • Coordinating several agents or coding command-line tools from one dashboard with scheduled work
  • Letting an agent answer or post in Discord, Slack, Telegram or WhatsApp
  • Switching between hosted models and local ones such as Ollama behind the same agents