Platform

Safety & Approvals

LiTT can edit files, run commands, and publish projects — real power, so it's gated by approvals. Here's what asks for your permission, and how to tune it.

The principle

Exploring is free; consequential actions ask first. LiTT can read your project, plan, and show you previews without bothering you. Anything that changes the world outside your workspace — deploying, publishing, or other irreversible steps — asks for your approval first. You stay the one who decides what ships.

Age and personal safety

LiTT is a general-audience service for people 13 and older and is not directed to children under 13. Never share passwords, payment-card numbers, government IDs, private keys, or another person's personal information in a prompt, upload, project, or public post. If you are under 18, involve a parent or legal guardian before using paid, business, publishing, or connected-account features.

• Use Stripe's hosted checkout for payments. LiTT does not ask you to paste a card number into chat or settings.
• Review generated code, images, and copy before publishing; AI output can be wrong or unsafe.
• Stop and report impersonation, harassment, sexual content involving minors, fraud, malware, or exposed credentials to support@litlabs.net.
• If a prompt asks for a secret or identity document, do not provide it unless the flow explicitly explains why it is required.

Approvals in Studio

• Deployment and publishing may present an approval step before anything goes live. Approve it and the deploy proceeds; decline or ignore it and nothing publishes.
• Plan review. For larger missions, LiTT shows its plan before acting — that's your cheapest moment to steer or stop.
• Session expiry. If your session expires mid-conversation, Studio tells you and asks you to sign in again rather than continuing in a broken state.

Permission modes in the CLI

The CLI lets you set how autonomous LiTT should be with --mode:

litt --mode plan    # plan only — LiTT proposes, you approve each step
litt --mode act     # act with approvals (default)
litt --mode auto    # proceed autonomously

• plan is the most cautious: LiTT lays out what it would do and waits.
• act (the default) does the work but still asks before consequential steps.
• auto proceeds without pausing — useful for trusted, repeatable workflows.

Separately, LITT_LOCAL_ONLY=1 blocks all remote and model use, forcing the CLI to stay fully local.

Models & routing

In Studio chat you choose how requests are handled:

• LiTT Auto routes your request automatically — you describe the goal and LiTT picks how to handle it.
• Specific agents and models let you take control: pick a specialist agent, and pick the underlying model — including your own API keys — with live provider health (available, degraded, locked) shown in the selector.

If a provider fails, Studio reports the failure openly instead of silently falling back — see Troubleshooting.

Best practices

• Review the plan before big missions — steering early is cheaper than reverting late.
• Check the preview before approving a deploy. “Done” means LiTT finished its steps; the preview is the source of truth.
• Use the activity timeline to audit what changed and what ran, especially before publishing.
• Keep --mode auto for workflows you've already watched succeed, not for first-time experiments.