Start here

Installing GUS, in plain English

From nothing to a working agent in about ten minutes, most of it waiting for a download. You do not need to be a programmer.

What you need

On a Mac, you can also install the native GUS.app instead — see the Mac app page.

Step 1 — Start GUS

Open a terminal on the computer and run:

git clone https://github.com/startekenterprises-ai/gus.git gus && cd gus
npm install && npm run build && npm run pack
docker compose -f packages/sandbox/docker-compose.yml up --build -d

That is the only time you need the terminal. The last command builds two boxes and starts them:

Step 2 — Open the control panel

In your web browser, go to http://127.0.0.1:8731. That address means "this same computer" — the panel is not open to the internet.

The panel asks for an operator token — a password GUS made up for you on first start. It lives in a file named panel-token. Read it with:

docker compose -f packages/sandbox/docker-compose.yml exec control-plane cat /state/panel-token

Copy it into the panel. Keep it like any password — whoever has it can approve actions and see your spending ledger (they still can't see your stored secrets; nobody can, including the agent).

Step 3 — The five-step setup wizard

  1. Connect Agent-Shield. Agent-Shield is the security scanner that checks everything the agent reads and does. This step creates a free key for you automatically and locks it in the vault. One click.
  2. Connect Telegram (optional but recommended). Create a bot by talking to Telegram's @BotFather (it takes a minute), then paste the bot token into the panel — never into a chat. From then on, approval requests and the stop button reach your phone. Honest fine print: Telegram chats are not end-to-end encrypted, so GUS only ever sends approvals and summaries there — never secrets.
  3. Pick a brain. Choose which AI model service the agent uses: a Google Gemini key (the recommended starter — a free key from Google AI Studio is enough to begin), your own Anthropic or OpenAI key, a model running on your own machine (Ollama), or a starter gateway if one is offered. Keys go into the panel and straight into the encrypted vault. One honest catch: on Google's free tier, Google may use what you send to improve its models — a paid Gemini key removes that.
  4. Set your spending caps. Pre-filled with starter limits — $2 per task, $10 per day, $100 per month. These are hard ceilings enforced outside the AI: when the money runs out, the work stops, full stop. Change them whenever you like, here in the panel.
  5. Run your first job. GUS runs a short read-only digest so you can watch a job go from start to a signed "receipt" — proof of what happened, what it cost, and that a checker (not the agent itself) verified the result.

Day-to-day

Updating

Updates arrive as signed releases. GUS checks the signature, tries the new version against your recorded automations before switching to it, and rolls back by itself if anything fails. You get a message before and after. Your data — the ledger, the vault, your approval history — is never part of a rollback; only code and settings are. On the Mac app, updating is one click — see Mac app.

If something goes wrong