Declarative orchestration for AI agent teams. Define agents in YAML, connect them over an embedded NATS message bus, and deploy anywhere — from a laptop to an air-gapped data center. One binary, no Docker, no Python dependencies.
How it works
Agents are defined in YAML with typed capabilities. Each agent gets a sidecar that handles heartbeats, capability registration, and message routing — agents just implement HTTP handlers. The sidecar auto-generates LLM tool definitions from capability schemas, so agents discover and invoke each other without hard-coded integrations.
Features
- YAML-driven agent definitions with typed capability schemas
- Embedded NATS message bus for inter-agent communication
- Firecracker microVM isolation on Linux (KVM), process backend on macOS
- Homebrew installable —
brew install brmurrell3/tap/hive - Built in Go