genai-io
GenAI Lab is an open-source AI infrastructure organization focused on agent software written in Go. Its flagship San project is a lean, fast, provider-agnostic terminal agent runtime, complemented by a Go LLM/agent SDK and related infrastructure such as an LLM gateway.
At a Glance
- Software developers and engineering teams
- Builders of coding agents and terminal automation
- Teams operating CI, containers, cloud VMs, edge devices or air-gapped environments
- AI/ML researchers and evaluation practitioners
- +1 more
AI Tools by genai-io
(1)
SAN
Open Source Terminal AI Agent
Discussions
No discussions yet
Be the first to start a discussion about genai-io
Latest News
San v1.23.0 released with workflows, Windows PowerShell support, auto-update controls, subagent graphs and autopilot panel
San v1.22.11 released with bounded resume reasoning, SDK protocol/data separation and inference failure events
San v1.22.9 released
San v1.22.8 released
Products & Services
Open-source terminal agent runtime delivered as a native Go binary. It supports multiple models and providers, personas, skills, subagents, plugins, MCP servers, autopilot, self-evolution, sessions, permissions, hooks, and a local inspector.
Go SDK underneath San, with a typed API across six LLM protocols, streaming, tool calling, structured outputs, typed errors, a 28-vendor catalog, and an event-based agent loop with hooks, parallel tools, sessions, and MCP.
Single-binary Go LLM gateway exposing Anthropic-compatible and OpenAI-compatible endpoints, unified model registry resolution, multi-provider routing, health checks, retry/failover, and streaming.
Open-source agents/persona bundles that can be used in the terminal on San.
Market Position
GenAI Lab positions San as a lean, open alternative to heavier terminal coding agents: its README benchmarks it against Claude Code and lists Aider and Continue as related projects. The differentiators claimed by the project are a roughly 12 MB native Go binary, zero runtime dependencies, roughly 0.01-second cold start, low memory/context overhead, provider and persona openness, and extensibility down to prompts, tools and extensions.
Founding Story
The project began with the vision of making an agent runtime small, fast, and open: a native Go binary with no Node.js or Python runtime dependencies, a small harness context, and user-controlled prompts, tools, providers, extensions, personas, and self-learning behavior. The San name is explained as 三 (three), referring to the three-step reason-act-observe loop and the Dao De Jing phrase ‘three begets the ten-thousand things.’
Business Model
Revenue Model
The projects are distributed as open-source software under Apache-2.0. The public materials document installation via binaries, Homebrew, Go modules and source builds, but do not describe a paid subscription, hosted service, API-usage charge or commercial pricing model.
Target Markets
- Software developers and engineering teams
- Builders of coding agents and terminal automation
- Teams operating CI, containers, cloud VMs, edge devices or air-gapped environments
- AI/ML researchers and evaluation practitioners
- Developers integrating multiple LLM providers through Go SDKs or compatible gateways
- Terminal software development and coding assistance
- Code review and repository exploration
- Long-running autonomous or autopilot engineering tasks
- CI runners, cloud VMs, containers, ARM edge devices and air-gapped hosts
- Specialized agents for ML research, security auditing, technical writing and other personas
- Multi-agent orchestration and MCP-based tool workflows