go-micro: an agent harness and service framework for Go that turns service endpoints into AI-callable tools

What it solves

Go Micro is an agent harness and service framework that simplifies the process of building and operating AI agents. It treats agents as distributed systems, providing the necessary runtime infrastructure—tools, memory, guardrails, and service discovery—so developers can focus on the agent's logic rather than the operational overhead of connecting models to real-world tools and services.

How it works

The framework integrates AI capabilities directly into a Go-based service architecture. It allows developers to write standard Go services where endpoints automatically become AI-callable tools. Agents are implemented as services with an Agent.Chat endpoint, enabling them to discover and use tools via a service registry.

Key mechanisms include:

• Agent Harness: Provides a pluggable runtime for models, durable store-backed memory, and tool execution.
• Plan & Delegate: Built-in tools that allow agents to record multi-step plans and hand off subtasks to other specialized agents via RPC.
• Gateways: An MCP gateway exposes services as tools, and an A2A gateway makes agents reachable to other frameworks using the Agent2Agent protocol.
• Durable Flows: Event-driven workflows that can be checkpointed to survive crashes.

Who it’s for

Go developers building AI agents that need to operate complex systems, manage state across turns, and interact with other agents or external services in a production environment.

Highlights

• Automatic Tooling: Service endpoints are automatically converted into AI tool schemas based on doc comments.
• AI-Driven Generation: A CLI that can generate entire services and handlers from a natural language prompt.
• Interoperability: Native support for the Model Context Protocol (MCP) and Agent2Agent (A2A) protocols.
• Durable Memory: Store-backed conversation memory that supports summarization for long-running interactions.
• Monetization: Support for the x402 payment standard for autonomous agent-to-agent tool payments.
• Multi-Provider: Compatible with seven major LLM providers including Anthropic, OpenAI, and Gemini.