Rune

Early development — APIs and features may change. We welcome feedback and contributions.

Production-grade AI agent runtime written in Rust. Rune handles the full lifecycle of AI agents: packaging, deploying, scheduling, executing, networking, and running them at scale — across Docker and WASM backends.

Key Features

• Multi-provider LLM — Anthropic, OpenAI, Google Gemini, GitHub Copilot, with automatic fallback
• MCP support — agents consume external MCP servers as tools; Rune exposes all deployed agents as an MCP server at POST /mcp
• Multi-backend — Run agents as WASM modules, Docker containers, or Kubernetes workloads
• Agent-to-Agent (A2A) — Agents can call each other via the Google A2A protocol over JSON-RPC
• Workflows — DAG-based multi-agent pipelines with parallel execution and conditional steps
• Compose — Deploy multiple agents together with dependency ordering
• Clustering — Optional Raft consensus for multi-node high-availability deployments

Quick Links

• Installation — Build and install the Rune CLI
• Quick Start — Deploy your first agent in minutes
• First Agent — Build an agent with a custom tool from scratch
• MCP — Using Rune with Claude Desktop, Cursor, and external MCP servers
• Examples — Browse all example agents
• CLI Reference — All rune commands
• HTTP API — Gateway endpoints
• Architecture — System design and crate map

Project Layout

Each agent is defined by a Runefile — a single YAML file combining identity, runtime, and model configuration:

my-agent/
├── Runefile        # identity, instructions, runtime, models
└── tools/          # optional custom tool implementations
    ├── my_tool.yaml
    └── my_tool.py   # or .js, .wasm, Dockerfile

For multi-agent stacks, a rune-compose.yml deploys everything together. See examples for complete samples.