Intent Composer

Turn a loose request into a brief you can see

The model reads your request, shows what it inferred, names what's ambiguous, and waits for you to steer — before it writes a word.

Awaiting request
1 · The request
As loosely as you'd actually type it

Most prompts are under-specified. Instead of guessing silently and generating confidently-wrong output, Steer parses the request first.

2 · Parsed intent

Two kinds of gaps, surfaced separately: things the model assumed (editable), and things it won't guess (you decide).

Inferred assumptions the model guessed these

Ambiguities the model refused to guess

3 · Generate
What the model will assume

A live read of the brief — updates as you edit an assumption or resolve a fork. This is the contract before any text is written.

Parse a request to see the brief.

Ambiguities resolved0 / 0
Decision log

Every steer is recorded — the brief is auditable, not vibes.

[--:--] waiting for a request…

Prototype · synthetic data. Theme follows the case-study toggle (light / dark). Try: parse the request, edit an assumption, resolve the forks, then generate — and compare against Generate blind.