Stanford Agent is a research-grade AI orchestration platform powered by arena.ai. Deploy, monitor, and collaborate with intelligent agents — from your phone, desktop, or anywhere in the world.
SaferSkills independently audited stanford-agent (Agent Skill) and scored it 100/100 (green). The audit ran 55 deterministic rules across Security, Supply Chain, Maintenance, Transparency, and Community; it found 0 high-severity and 0 lower-severity findings. The full rule-by-rule trace and per-finding evidence are below. Free, methodology-open.
Findings & checks · 0 flagged
Every scanned point with the score it earned and what moved between them.
First recorded scan — no prior version to compare against.
The primary manifest — the file an agent reads to learn what this artifact does.
<p align="center"> <img src="assets/logo.png" width="90" alt="Stanford Agent" /> </p>
<h1 align="center">Stanford Agent</h1>
<p align="center"> <b>Research-grade AI agent orchestration — backed by <a href="https://x.com/arena">arena.ai</a></b><br/> Run, monitor, and collaborate with AI agents from your phone, desktop, or CLI. </p>
<p align="center"> <a href="LICENSE"><img src="https://img.shields.io/badge/license-MIT-blue.svg" alt="MIT License"/></a> <img src="https://img.shields.io/badge/backed_by-arena.ai-FF6B35" alt="Backed by arena.ai"/> <a href="https://stanfordagent.cloud"><img src="https://img.shields.io/badge/web-stanfordagent.cloud-111111" alt="Website"/></a> </p>
<p align="center"> <a href="https://stanfordagent.cloud">🌐 Website</a> · <a href="docs/getting-started.md">📖 Docs</a> · <a href="https://x.com/stanford_ee">𝕏 @stanford_ee</a> · <a href="https://x.com/ml_angelopoulos">𝕏 Founder</a> · <a href="https://x.com/arena">𝕏 arena.ai</a> </p>
Stanford Agent is an open-source terminal daemon that lets you spawn, monitor, and collaborate with AI coding agents across any number of machines — all from a single unified interface. It runs entirely on your hardware; the optional arena.ai encrypted relay lets you reach your agents from anywhere without exposing local ports.
It supports Claude Code, OpenAI Codex, OpenCode, Cursor, GitHub Copilot, arena.ai Pi, and 26+ other providers. Each agent runs as its own native process using its own API key — Stanford Agent never modifies or wraps agent behavior.
# Install globally
npm install -g @stanford-agent/cli
# Start the daemon + open dashboard
sagent
# Or run headlessly (no TUI)
sagent daemonRequirements: Node.js ≥ 18, Git ≥ 2.38 (for worktree support)
# Spawn a new agent task
sagent run "build a REST API for student research data"
# List running agents
sagent ls
# Attach to a running agent (stream its output)
sagent attach abc123
# Stop an agent
sagent stop abc123
# Tail logs
sagent logs abc123 --follow
# Voice control (local STT/TTS)
sagent voice
# Show config
sagent configExample session:
❯ sagent run "build REST API for student research data"
🤖 Spawning agent on arena-cluster-1…
📦 Worktree: feat/research-api
🔗 Preview: web.feat-research-api.myapp.localhost
❯ sagent ls
ID BRANCH STATUS
abc123 feat/research-api ● running
def456 fix/auth-token ● running
ghi789 feat/ml-pipeline ⏸ paused
❯ sagent attach abc123
✓ Compiled routes/research.ts in 290ms
✓ Tests passed (18/18)
# Open PR? [y/n]| Feature | Description |
|---|---|
| 🤖 Multi-Agent Orchestration | Run dozens of agents in parallel across sessions, branches, and machines |
| 📱 Universal Access | Full-parity mobile, web, and desktop apps via arena.ai relay |
| 🔒 E2E Encrypted Relay | Your code never leaves your machine; arena.ai cannot read your traffic |
| 🎙️ Local Voice Control | Fully local speech-to-text and TTS — no audio sent to external servers |
| ⚡ Zero Port Conflicts | Each agent worktree gets a unique URL auto-generated from the branch name |
| 🛠️ Fully Scriptable CLI | Every UI action available from the terminal; build loops, schedules, and pipelines |
| 🔌 30+ Providers | Claude Code, Codex, OpenCode, Cursor, Copilot, arena.ai Pi, and more |
Stanford Agent uses each provider's native harness — your subscriptions, skills, config, and MCP servers keep working.
| Provider | Status |
|---|---|
| Claude Code | ✅ Stable |
| OpenAI Codex | ✅ Stable |
| OpenCode | ✅ Stable |
| Cursor | ✅ Stable |
| arena.ai Pi | ✅ Stable |
| GitHub Copilot | ✅ Stable |
| Aider | ✅ Stable |
| Continue | ✅ Stable |
| Cody (Sourcegraph) | 🧪 Beta |
| _+ 22 more_ | 🧪 Beta / Community |
See docs/providers/ for per-provider configuration.
┌─────────────────────────────────────────────────┐
│ sagent daemon │
│ ┌──────────┐ ┌──────────┐ ┌──────────────┐ │
│ │ Agent 1 │ │ Agent 2 │ │ Agent N… │ │
│ │ (claude) │ │ (codex) │ │ (opencode) │ │
│ └──────────┘ └──────────┘ └──────────────┘ │
│ │ │ │ │
│ └──────────────┴──────────────┘ │
│ │ │
│ ┌──────────────────┐ │
│ │ relay server │ │
│ │ (E2E encrypted) │ │
│ └──────────────────┘ │
└─────────────────────────────────────────────────┘
↕ arena.ai encrypted relay
┌─────────────────────────────────────────────────┐
│ Remote clients (optional) │
│ mobile app │ web dashboard │ CLI │
└─────────────────────────────────────────────────┘The daemon runs entirely on your machine. The arena.ai relay is an optional encrypted tunnel — useful when you need to reach your agents from outside your local network. End-to-end encryption means arena.ai sees only ciphertext.
Stanford Agent stores config at ~/.config/stanford-agent/config.json. Edit it directly or use sagent config:
sagent config set relay.enabled true
sagent config set relay.token YOUR_ARENA_TOKEN
sagent config set voice.provider whisper.cpp # local STT
sagent config set tts.provider piper # local TTSA full config reference is in docs/cli-reference.md.
Is Stanford Agent free? Yes — fully free and open source (MIT). You need your own API credentials for the AI providers you use. The arena.ai relay is included for research preview participants.
Does my code leave my machine? No. Agents run locally using their own APIs. Remote access uses end-to-end encryption — arena.ai cannot read your traffic.
Do I need the desktop app? No. Run sagent daemon headlessly and connect via web, mobile, or CLI from anywhere.
How does voice control work? Voice runs fully local by default using whisper.cpp for STT and Piper for TTS. Cloud providers are configurable for higher accuracy.
We welcome issues, discussions, and pull requests.
git clone https://github.com/StanfordAgent/stanford-agent.git
cd stanford-agent
npm install
npm run dev # watch modeSee CONTRIBUTING.md for guidelines.
MIT © 2026 Stanford Agent contributors. See LICENSE. Backed by [arena.ai](https://x.com/arena) · Research-grade AI infrastructure.
~30 seconds. Free. No account. Every finding cites a rule and a line of evidence.