SwarmClaw

SwarmClaw is a self-hosted, open-source AI agent runtime that lets you build, orchestrate, and run autonomous multi-agent systems with durable memory, runtime skills, delegation, and scheduled execution. It supports 23+ LLM providers including Claude, GPT, Gemini, Ollama, OpenRouter, and more, and is natively compatible with the Model Context Protocol (MCP). Whether you're running a personal assistant, a virtual company org chart, or a distributed OpenClaw fleet, SwarmClaw provides the control plane to wire agents together.

• Multi-provider support — Connect 23+ built-in providers (Anthropic, OpenAI, Gemini, Ollama, DeepSeek, Groq, Mistral, xAI, Fireworks, and more) plus custom OpenAI-compatible endpoints.
• MCP-native architecture — Connect any Model Context Protocol server via stdio, SSE, or streamable HTTP and inject its tools into agents alongside built-ins.
• Delegation and orchestration — Delegate work to subagents, CLI backends (Claude Code, Codex, Gemini CLI, Cursor, etc.), or orchestrator agents with durable branching, parallel joins, and restart-safe run state.
• Autonomy and scheduling — Heartbeat loops, scheduled jobs, background execution, supervisor recovery, and agent wakeups keep agents running without human intervention.
• Hybrid memory — Graph traversal, journaling, durable documents, project-scoped context, reflection memory, communication preferences, and open follow-up loops.
• Structured Sessions — Reusable bounded runs with templates, facilitators, participants, durable transcripts, operator controls, and a visual protocol builder.
• Conversation-to-skill drafts — Draft a reusable skill from any real chat, review it, then approve it into the shared skill library.
• Crypto wallets — Link Solana and Ethereum wallets to agents for balance checks, transaction simulation, approval-gated execution, and USDC payouts via SwarmDock.
• Connectors — Bridge agents to Discord, Slack, Telegram, WhatsApp, Teams, Matrix, OpenClaw, SwarmDock, and SwarmFeed.
• Desktop app and CLI — Install via one-click desktop installer (macOS, Windows, Linux) or globally via npm/yarn/pnpm/bun; runs on http://localhost:3456.
• Hosted deployment — Deploy using provider-ready render.yaml, fly.toml, or railway.json files from the repo root with the published GHCR image.
• OpenTelemetry tracing — Opt-in OTLP trace export for chat turns, model streams, tool execution, and structured-session runs.