A rehearsal room, and it says so.
RostraLab builds synthetic opponents for private preparation. That is a tool with obvious ways to be misused, so here is what it does and what it refuses.
Everything carries the label
Every debate room, every recording and every transcript is marked as an AI training simulation — on screen and in the files you download — in line with the EU's AI transparency rules. Nothing made here is meant to be published as a real recording.
No voice cloning
The AI opponent argues like your opponent, from their public record, but speaks with an original, generic voice. We do not clone, imitate or approximate the voice of a real person, and we do not accept voice samples of one.
Built from the public record
Opponent personas are built from public statements, votes, broadcasts and press — what a person said in public, with a link to where they said it. Not private data, not leaked material, nothing a citizen could not look up themselves.
Interpretation is approved by a person
What an AI reads is split in two: what somebody verifiably said, and what our tools inferred from it. The inference is withheld from every persona until somebody on your team approves it. An AI opponent never argues from something nobody signed off on.
What it is not for
It is a rehearsal room, not a content factory. It is not built to generate campaign material in somebody else's voice, to produce a recording of a person saying what they never said, or to publish anything at all — and a customer who wanted it for that would be the wrong customer.
Ask us the hard version
If your ethics board wants the long answer, we will give it.
Request access →