Trust

Effective date: 25 September 2026

Applies to Unl, your why agent, on unlimitless.ai and api.unlimitless.ai

Your agents move fast. Your decisions stay yours. Agents act more on their own every month, and that is good. Unl is how you let them run further without losing the thread. It keeps what you decided and why, and hands each agent the part that bears before it acts. The deciding stays with you.

What Unl does

It finds what bears and hands it over. When your AI or your agent is working, Unl finds the decisions that bear on the work in front of it and puts them, with their reasons, into the model’s context. The model does the thinking.

It notices what looks like a decision. When something you say looks like a new decision, or a change to one you hold, Unl tells your AI, and your AI asks you in the same conversation. Nothing is kept until you say so, in your own words.

It speaks before a decision is crossed. When an agent is about to act against something you decided, Unl puts the decision and its why in front of it first. This is rolling out now.

What Unl never does

It never decides for you, and it never makes something one of your decisions. It never writes to your world on its own, and it never reaches into your model, your agent or your tools to inspect or steer them. It works only on what it is handed.

Where authority sits

With you, by construction. Reads flow freely. Writes do not. Every write carries a single-use token that exists only because you connected, and it is checked at one place that every write passes through. So a model cannot decide to write, and neither can a message, a web page or a tool result that happens to contain instructions. There is no machine path to your decisions.

Reasons, not rules. A rule without a why reads as a yes to everything it did not name. A decision with its reason tells an agent what it covers, which is why your agents can be trusted with more of the work.

What is kept, and for how long

What you save is kept for as long as you have an account. The decisions you settle, the reasoning behind them, the targets you commit to and the thoughts you keep. Nothing is overwritten. A position you change is superseded and the earlier one stays readable, so you can always walk back to what you thought before and why you moved.

What your tools return passes through. When your AI asks one of your tools through Unl, the reading travels to your AI in that moment and is not kept.

What Unl told your AI is noted for 30 days. Each time Unl hands your AI your decisions, it notes which decisions went to which tool, model and agent. Ask any connected AI what Unl told it and it can show you. The note holds no words from your conversation and changes no decision.

Product analytics are technical events, never your content, and are kept in de-identified form.

Is anything used to train AI

No. We never train AI models on your content, and neither may the providers who help us run Unl. Each sees only what its job needs. Clerk handles sign-in, with your name, email and sign-in identifiers. Anthropic’s Claude API does limited housekeeping on saved items, such as proposing a summary for you to confirm, under terms that prohibit training on the data. PostHog, hosted in the EU, receives technical events and pseudonymous identifiers, never your content. Slack receives feedback you choose to send us. Database hosting, application hosting and error monitoring complete the list, and error reports are set up so they cannot contain your content. The full, current list is yours on request at privacy@unlimitless.ai.

How to export everything

From your portal, whenever you like, as machine-readable JSON. It carries every decision with its reasoning, what it replaced and when. You can also ask us at privacy@unlimitless.ai for everything we hold about you.

How to delete everything

From your portal settings you can schedule deletion yourself. Your account then waits for 30 days, during which you can cancel and your export stays available. After that your content and personal data are deleted for good, and any copies in encrypted backups clear within a further 30 days. The detail is in Privacy, section 7.

What this is, and what it is not

The labs are working on control at the level of the frontier. Unl works at the level of the person, over the agents you actually run. It does not claim to make AI safe. It keeps your reasons in front of your agents and the deciding in your hands, and we measure how often a fleet crosses a decision with Unl and without it.

For how this is built and checked, see Security.