Open agent infrastructure, written in Go.

Two repositories carry the work: San, a terminal agent runtime you can open all the way down, and sdk-go, the LLM client and agent SDK it runs on.

⚡ ~0.01s cold start  ·  📦 ~12 MB single binary  ·  🪶 zero runtime deps

An open-source terminal agent runtime: one native Go binary, no Node.js and no Python. Everything the model touches — prompt, tools, providers, extensions — stays yours to change.

• Small — ~2.3k tokens of harness reach the model before your first message; the rest of the context window goes to your work.
• Fast — a full tool-use task returns in ~3.3s end to end. What you wait on is the model, not the client.
• Open — plug in any model, skill, subagent, or MCP server; write your own system prompt and autopilot goals; replay any run in san inspector.

brew tap genai-io/san && brew install san    # or: curl -fsSL https://raw.githubusercontent.com/genai-io/san/main/install.sh | bash

Website · Getting started · Docs · 简体中文

A Go SDK for large language models, in two packages — and the engine San itself is built on.

• pkg/ai — one model call. One typed API over five protocols: Anthropic Messages, Anthropic on Vertex AI, OpenAI Chat Completions, OpenAI Responses and Google Gemini. Streaming, tool calling, structured outputs, typed errors, a catalog of 27 vendors, and no ambient credentials.
• pkg/agent — the loop around it. Reason and act, everything as events, four hooks to refuse or rewrite a tool call, parallel tools, and sessions you can restore.

go get github.com/genai-io/sdk-go

Go reference · README · 中文文档

• personas — agents you can hire into your terminal, running on San
• ml-researcher — ML research engineer persona for experiments, training, and evaluation
• llm-gateway — universal LLM gateway with Anthropic- and OpenAI-compatible endpoints and multi-provider routing
• awesome-agent-evals — curated evaluation resources for AI agents: platforms, frameworks, benchmarks, methodology
• handx — reach your dev machine's tmux sessions from your phone
• spec — GenX multi-agent system specification
• bell · orchestrator — event bus and lifecycle management for multi-agent orchestration