How it works: ask, predict, act

intenthybrid AI runs the same three steps every time, through a private channel. This is the core loop of the agent. Understanding it makes everything else in the documentation easier to follow.

1. Ask

You bring the agent a problem in plain language. This happens in private chat, the core of the product: the conversation lives in your browser, and your prompt runs on infrastructure you control. Nothing about the question is sent to an intenthybrid AI server.

Ask answers: what do you want solved?

2. Predict

The agent reasons over the problem and produces a prediction, not just a restatement of the inputs. It interprets the situation and estimates what happens next: what is likely to break, what unwinds, where the risk is heading. The output is a plain-language result you can act on.

Predict answers: what does this mean, and what happens next?

3. Act

A prediction is only worth having if you can do something with it. intenthybrid AI gives you three ways to act on the same result:

• Get a result back in chat, then decide for yourself.
• Set an alert, so you are notified the moment a condition you care about is met.
• Arm a scoped agent, so the agent acts on its own prediction inside hard limits you define, whether or not you are watching.

Act answers: what happens now, and who does it?

Privacy wraps the whole loop

Ask, predict, and act all run through the private channel. Your conversation and history stay local, and your prompts are not logged to a third party. Privacy is not a fourth step, it is the condition that holds across all three. See The core, the privacy model.

The same loop, applied

This loop is general. In the flagship application, perp-market risk, ask becomes a question about a market, predict becomes a read of crowding and what would unwind, and act becomes an alert or a guardrailed agent that reduces exposure. See The runtime spine for the engineering view, and Application: perp-market risk for the use case.