Give your AI
your computer.

Give it a Mac, Linux box, or Docker container. It takes over the screen, keyboard, filesystem, and terminal — then works around the clock on whatever you point it at. Talk to it through chat or Slack. Watch it work live. Take over via VNC when it needs a human.

⚠️ Gives Claude full machine access. Run on a dedicated machine.
terminal
git clone https://github.com/prestoj/relaygent.git
cd relaygent
./setup.sh
relaygent start

What it does

🔁

Runs indefinitely

When one session fills its context window, the agent writes a detailed briefing and a fresh session picks up exactly where it left off. Hours, days, weeks.

🖥

Controls your screen

Screenshots, clicks, typing, scrolling, accessibility trees, app launching, browser navigation. Full computer use on macOS and Linux.

🧠

Remembers everything

A git-tracked knowledge base persists across sessions. The agent reads its predecessor's notes, updates them, and leaves better ones for the next session.

💬

Chat, Slack, and file sharing

Message your agent through the dashboard or Slack. Share files both ways. Watch it work in real time with live activity streaming. Check in from your phone.

🖳

VNC when it needs you

Take over the screen via VNC when the agent hits a captcha, needs approval, or gets stuck. Help it, then hand control back. Human-in-the-loop on your terms.

🐳

Docker in 60 seconds

docker compose up -d and open localhost:8080. Full desktop environment, VNC access, and all services — no host setup required.

See it in action

How the relay works

1

Start a session

The harness launches a Claude Code session with a system prompt, handoff notes from the previous session, and your intent file.

2

Agent works autonomously

Claude reads files, writes code, runs commands, uses the screen — whatever the task requires. A PostToolUse hook injects time, notifications, and context usage after every tool call.

3

Context fills up

At ~85% context, the agent wraps up: writes a detailed handoff, updates persistent state, and commits the knowledge base.

4

Successor continues

The harness spawns a fresh session that reads the handoff and continues immediately. No information lost, no manual intervention.

Requirements

macOS or Linux

macOS 13+ or Ubuntu 22.04+. Computer use via Hammerspoon (macOS) or xdotool (Linux).

Claude Code

npm install -g @anthropic-ai/claude-code with an active Anthropic subscription.

Node.js 20+ & Python 3.9+

For the dashboard, services, and harness. Setup handles the rest.

Common questions

Does this cost money?

Relaygent is free and open source. You'll need an active Anthropic subscription since it uses Claude Code. There's no additional cost beyond your existing Claude plan.

Is it safe to run?

Relaygent gives Claude full access to your machine — files, terminal, browser, screen. There is no sandbox and no kill switch beyond stopping the process. We strongly recommend running it on a dedicated machine or VM, not your primary computer.

Does it run 24/7?

Yes, if you want it to. Between active work the agent sleeps at zero token cost. It wakes on messages, reminders, or notifications. When a session fills its context window, a fresh session picks up automatically — no manual intervention needed.

What can I actually use it for?

Anything you'd ask a skilled developer or assistant to do on your machine: write and run code, manage files, browse the web, monitor systems, respond to messages, research topics, maintain a knowledge base. You define the priorities in an intent file.

Which AI models does it use?

Relaygent is built around Claude Code, so it works with the latest Claude models. You need an active Anthropic subscription (Max or Team plan). There's no additional API cost — it runs through your existing Claude Code access.

Can I see what it's doing?

Yes — the built-in hub dashboard shows a live activity feed, session history, chat interface, and knowledge base browser. You can message the agent at any time and it will respond. Since the agent uses the desktop, you'll typically access the hub from another computer on the same network — just navigate to the machine's IP on port 8080.

Can I run multiple agents?

Yes, and we recommend it. Two agents working together review each other's code, split tasks, and catch mistakes — they're more than the sum of their parts. Give each agent its own machine (or VM), its own email and GitHub identity, and connect them via Slack and Linear for coordination. See the Recommended Setup guide for details.

What integrations does it support?

Natively: Slack (real-time via Socket Mode), Gmail (OAuth2), GitHub (notifications + CLI), and Linear (task management). For anything else — Microsoft Teams, ClickUp, Jira, Notion — ask your agent to build the integration. It can write custom MCP servers and configure webhooks for any API.

Get started in 60 seconds

terminal
git clone https://github.com/prestoj/relaygent.git
cd relaygent
./setup.sh
relaygent start
View on GitHub