An Agent Harness for Go
Build agents, services, and workflows on one runtime.
curl -fsSL https://go-micro.dev/install.sh | sh
Build agents, services, and workflows on one runtime.
curl -fsSL https://go-micro.dev/install.sh | sh
An agent harness and a service framework in one — agents, services, and flows on the same runtime, with the production pieces agents need.
A model, memory, tools, plan/delegate, guardrails, and execution middleware around every agent.
Build services in Go or generate them from a prompt. Their endpoints become typed tools automatically.
Use fixed, checkpointed code paths for deterministic work; hand off to agents when the path is dynamic.
Every service endpoint is automatically an AI-callable tool via the Model Context Protocol.
Every agent is reachable over the Agent2Agent protocol — discovered and called by agents on any framework.
All abstractions are Go interfaces. Swap any component without changing your code.
Tell it what you need. The AI designs services, generates an agent, and drops you into an interactive console. The harness gives that agent tools, memory, guardrails, and workflows so it can operate across services.
$ micro run --prompt "task management system" > Create a project called Launch, add tasks, assign to Alice
Agents need the same substrate services do: discovery, RPC, events, state, auth, observability, and deployment. Swap any component without changing your code — go from mDNS to Consul, or HTTP to gRPC, with a single option.
micro new scaffolds a service or an agent — every endpoint is automatically an MCP tool. micro run starts everything with hot reload, an API gateway, and an interactive console, and micro chat talks to your agents and services from the terminal.
Go Micro is supported by companies building the future of AI infrastructure.
Want to support Go Micro and put your logo here? Or running it in production and need a hand?
23,000+ stars on GitHub. Production-ready. Apache 2.0 licensed.