Human Start Here

ORBIT is a checkpoint for agent actions.

If an AI agent is about to do something with real consequences, ORBIT tries to make the action visible, governed by your policy, and auditable afterward. It is not magic, and it does not see every possible thing a computer can do. It governs the tool calls and adapters you install it on.

The plain-English model

Think of ORBIT as a permission desk between an agent and the tools it has been wired to use. Routine work can go through. Risky work can ask you first. Work you never want delegated can be blocked. Governed decisions are recorded as receipts when the hooked path reaches ORBIT.

The important boundary is honest: ORBIT can only decide on actions its hook or adapter observes. If a tool is not connected to ORBIT, ORBIT cannot claim to have governed it. The docs call out those coverage states instead of pretending otherwise.

Your first hour

  1. Sign in and verify the phone number that should receive approval prompts.
  2. Copy your API key from /keys. It is shown once.
  3. Install Guardian using the manual install guide.
  4. Run orbit-hook whoami to confirm the hook can identify your account.
  5. Ask your agent to do one safe thing, then check /audit for the record.

The Quickstart has the same path with commands and troubleshooting links.

Copy this prompt if you want your AI to help

This keeps the helper useful without letting it quietly turn into a forklift in a china shop.

Ask your AI to do this

Risk label: Human-guided ORBIT onboarding

Stop gate: Stop before exposing API keys, installing software, changing policy, approving actions, mutating production, or spending money unless I explicitly approve that step.

You are helping me understand and set up ORBIT. Read /docs/human-start-here and /docs/quickstart first. Explain ORBIT in plain language, name what it can and cannot govern, and help me follow the first-hour checklist. Stay read-only until I approve a specific action. Do not expose API keys, change policy, install software, contact support, mutate production, or claim ORBIT governed paths it did not observe. Return the exact evidence I should see in /audit after one harmless governed action.

What happens on your phone

When a rule says “Ask me first,” ORBIT pauses the action and sends a text to your verified phone. The text says what the agent wants to do, why it matters, what ORBIT observed, and how to reply.

Reply Y to allow once, N to block, or 10M, 1H, or 2H for a time-bounded standing approval when that option is offered. If you do nothing, the request times out and the action is denied rather than silently allowed.

What proof you get

A receipt proves that ORBIT made a decision about the observed action under a named policy at a particular time. It does not prove that the underlying work was wise, harmless, or complete. Some receipts include a reported outcome from the agent; stronger outcome verification depends on the adapter and receipt tier.

Download receipts as .orbitproof files and verify them at /verify. For the deeper receipt model, read Verifying receipts.

Three normal examples

  • Allowed: the agent lists project files. ORBIT records the observed read-only action and lets it continue.
  • Ask first: the agent wants to run a broad delete or cloud command. ORBIT texts you before the action proceeds through the governed path.
  • Blocked: the agent tries a pattern your policy marks as never allowed. The agent gets a block verdict and the audit log explains the matched rule.