San is a terminal-native unified runtime for specialized agents — coding and beyond — built on five pluggable pillars: LLMs, search backends, personas, skills & extensions (skills, plugins, MCP servers, subagents), and a self-evolving agent that levels up as you work. Written in Go.

_{The name — San, written 三 ("three") and drawn ☰. From the Dao De Jing, 三生万物 — "three begets the ten-thousand things": one runtime that becomes any agent, running a three-step loop (reason → act → observe). The command stays san.}

• LLM providers — Anthropic, OpenAI, Google, DeepSeek, Moonshot, Alibaba, MiniMax, Z.ai (GLM), SenseNova, Mimo, Volcengine (Ark), Ollama (local); swap via /model.
• Search backends — Exa, Tavily, Brave, Serper; swap via /search.
• Personas — Markdown identities scoped to user or project; swap via /identity (details).
• Skills & extensions — Claude Code skills, plugins, and MCP servers run unmodified; sandboxed subagents; lifecycle hooks (shell, LLM, agent, HTTP); auto-loaded project memory.
• Self-evolving — every few turns a background reviewer distills your recent work into durable memory and reusable skills, so the agent levels up as you work. (Level 1 available; deeper levels on the way.)

• Runs anywhere — A single ~12 MB binary, zero runtime dependencies (no Node.js, no Python). Native Go: ~0.01s cold start, ~32 MB baseline; the same file runs unchanged on a laptop, an edge device, or a scratch container. Windows, macOS, and Linux builds in release artifacts (footprint · benchmark).
• Permission system — Mode-based access control: ask (default) and auto-accept, toggled with Shift+Tab. Subagents inherit permission gates for sandboxed execution (details).
• Event-driven coordination — Parallel subagents via a pub/sub hub (architecture).
• Session persistence — Auto-save, resume (--continue, --resume), fork (/fork), and automatic context compaction (/compact).
• Cost tracking — Per-message and per-session token costs across all providers.
• Theme & appearance — Switch TUI color themes via the /config Appearance panel; "auto" matches your terminal.
• Prompt prediction — Speculative completion of likely next prompts to cut latency.
• Session inspector — Local web UI for transcript replay, system-prompt forensics, and live-tail of active sessions (san inspector).

macOS / Linux

curl -fsSL https://raw.githubusercontent.com/genai-io/san/main/install.sh | bash

Windows (PowerShell)

irm https://raw.githubusercontent.com/genai-io/san/main/install.ps1 | iex

Re-run to upgrade.

# macOS / Linux
curl -fsSL https://raw.githubusercontent.com/genai-io/san/main/install.sh | bash -s uninstall

# Windows (PowerShell)
& ([scriptblock]::Create((irm https://raw.githubusercontent.com/genai-io/san/main/install.ps1))) uninstall

go install github.com/genai-io/san/cmd/san@latest

git clone https://github.com/genai-io/san.git
cd san
go build -o san ./cmd/san
mkdir -p ~/.local/bin && mv san ~/.local/bin/

san                              # interactive
san "explain this function"      # one-shot
san -p "do something"            # non-interactive print mode (no TUI)
echo "data" | san -p "analyze"   # piped input in print mode
san --continue                   # resume latest session
san --resume                     # pick a past session
san --resume <session-id>        # resume specific session by ID
san --plugin-dir <path>          # load plugins from a specific directory

# Subcommands
san inspector                    # open session transcript viewer
san agent run --type Explore --prompt "find main.go"  # run a headless agent
san plugin list                  # list installed plugins
san plugin install <ref>         # install a plugin
san plugin enable <name>         # enable a plugin
san plugin disable <name>        # disable a plugin
san plugin uninstall <name>      # uninstall a plugin
san mcp add <name> -- <cmd>      # add an MCP server (stdio)
san mcp add --transport http <name> <url>  # add an MCP server (HTTP/SSE)
san mcp list                     # list MCP servers
san mcp get <name>               # get MCP server details
san mcp edit <name>              # edit MCP server config in $EDITOR
san mcp remove <name>            # remove an MCP server

For API keys, set the matching env var (see Credentials below) or paste when prompted on first launch. Full walkthrough: docs/guides/getting-started.md.

Config lives in ~/.san/ (user) and <project>/.san/ (project, overrides user). A SAN.md or CLAUDE.md at the project root is auto-loaded into the system prompt.

providers.json    # Provider connections and current model
settings.json     # Permissions, hooks, env, identity
skills.json       # Skill states
identities/       # Custom personas (see /identity)
skills/           # Custom skill definitions
agents/           # Custom agent definitions
commands/         # Custom slash commands
plugins/          # Installed plugins
projects/         # Session transcripts + indexes

settings.json       # Permissions, hooks, disabled tools
mcp.json            # MCP server definitions (team shared)
mcp.local.json      # MCP server definitions (personal, git-ignored)
identities/*.md     # Project-scoped personas (override user-level)
agents/*.md         # Subagent definitions
skills/*/SKILL.md   # Skills
commands/*.md       # Slash commands
plugins/            # Project-level plugins
plugins-local/      # Local plugins (git-ignored)

Compared with Claude Code v2.1.112 on Apple Silicon, same model (claude-sonnet-4-6):

Both tools have comparable features (hooks, skills, plugins, session, MCP, etc.). The performance gap comes from Go's native compilation, minimal architecture design, and lean prompt engineering — vs Node.js V8/JIT/GC runtime overhead.

See full details: docs/operations/benchmark.md

• Documentation Index — map of architecture, features, operations, and references
• Architecture — architecture entrypoint and reading order
• Package Map — package ownership and dependency boundaries
• System Prompt — Slot model, identity, skill/agent injection
• Subagents · Skills · Plugins · MCP
• Hooks · Permissions · Tasks
• Inspector — local web UI for transcript replay and debugging
• Per-package design under docs/packages/ — start at Package Index

• Claude Code — Anthropic's AI coding assistant
• Aider — AI pair programming in terminal
• Continue — Open-source AI code assistant

Two ways in — WeChat for the Chinese community, Slack for everyone else:

关注公众号「极客外传」· 回复 san 或 三 入群

Scan or join our Slack

Contributions welcome! See CONTRIBUTING.md for guidelines.

Apache License 2.0 - see LICENSE for details.