convez.ai
MCP ServerAutomaçãoAvançado
genai-io

SAN — Runtime de Agentes IA para Terminal

genai-io/san

Runtime terminal para agentes IA especializados com arquitetura plugável

Ver no GitHub
BomPluginScore · de 0 a 100

Runtime open-source unificado para agentes IA especializados, executado nativamente no terminal. Suporta múltiplos provedores de LLM, backends de busca, personas customizáveis e extensões (skills, plugins, MCP servers). Escrito em Go com zero dependências externas, roda em qualquer lugar com ~12 MB de binary.

Instalação

$npx convez-ai genai-io/san

1 comando no terminal: detecta o formato do repositório (marketplace ou plugin avulso) e instala via Claude Code. Requer Node 18+ e o Claude Code instalado.

Métricas

Estrelas
147
Forks
36
Acompanhando
2
Saúde
87%
Último commit
há 1 dia
Versão
v1.25.3
Issues abertas
3
Contribuidores
14

Casos de uso

  • Automação de tarefas de desenvolvimento no terminal
  • Execução de agentes IA locais com múltiplos modelos LLM
  • Integração de skills, plugins e MCP servers em workflow
  • Desenvolvimento de agentes customizados com personas e memória persistente
  • DevOps e automação de infraestrutura com agentes IA

Veredito

SAN é uma solução sólida e bem arquitetada para quem quer agentes IA no terminal com controle total. A abordagem modular (LLMs plugáveis, personas, skills) e o footprint mínimo (Go puro, zero dependências) o tornam atrativo para ambientes restritos. Recomendado para desenvolvedores e arquitetos que precisam de agentes customizáveis e self-evolvable.

README

A tradução para o português deste README ainda está na fila.

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.

Features

Open architecture

  • 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.)

Engineering

  • 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).

Installation

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.

Uninstall

# 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

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

Build from Source

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/

Usage

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
WhatHow
Pick / switch model/model — saved to ~/.san/providers.json
Cycle thinking budgetCtrl+T or /think (levels vary by provider)
Toggle permission modeShift+Tab (ask · auto-accept)
Search / identity / memory/search · /identity · /memory
Skills / agents / tools/skills · /agents · /tools
Plugins / MCP / config/plugin · /mcp · /config
Session / loop / misc/fork · /compact · /loop · /glob · /init · /clear
All slash commands/help
Send · newline · stopEnter · Alt+Enter · Esc
Expand tool · cancel · exitCtrl+O · Ctrl+C · Ctrl+D

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.

Configuration

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.

ServiceVariable
Anthropic (Claude)ANTHROPIC_API_KEY or Vertex AI
OpenAI (GPT, o-series, Codex)OPENAI_API_KEY
Google (Gemini)GOOGLE_API_KEY
DeepSeek (DeepSeek V4)DEEPSEEK_API_KEY
Moonshot (Kimi)MOONSHOT_API_KEY
Alibaba (Qwen)DASHSCOPE_API_KEY
MiniMaxMINIMAX_API_KEY
Z.ai (GLM)BIGMODEL_API_KEY
SenseNovaSENSENOVA_API_KEY
MimoMIMO_API_KEY
Volcengine (Ark)VOLCENGINE_API_KEY
Ollama (local)OLLAMA_BASE_URL (default http://localhost:11434/v1)
Exa searchnone (default)
Tavily searchTAVILY_API_KEY
Brave searchBRAVE_API_KEY
Serper searchSERPER_API_KEY

User-level (~/.san/):

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

Project-level (.san/):

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)

Benchmark: San vs Claude Code

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

MetricSanClaude CodeAdvantage
Download size12 MB63 MB (+ Node.js 112 MB)5x smaller
Disk footprint38 MB175 MB4.6x smaller
Startup time~0.01s~0.20s20x faster
Startup memory~32 MB~189 MB5.8x less
Simple task~2.4s / 39 MB~10.4s / 286 MB4.3x faster, 7.3x less memory
Tool-use task~3.3s / 39 MB~26.0s / 285 MB7.9x faster, 7.2x less memory

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

Related Projects

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

Community

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

Contributing

Contributions welcome! See CONTRIBUTING.md for guidelines.

License

Apache License 2.0 - see LICENSE for details.

Ver o original no GitHub

Bom para

Relacionados