An AI cofounder that remembers everything. And still says no.

Mission

The biggest leak in solo-founder productivity is not lack of ideas. It is having no one to say no when the data says no. General AI tools either agree with you out of politeness or hallucinate evidence. We built FounderDive to be neither: an AI cofounder that scans live market data, scores your idea across six dimensions, and refuses to give you a third Deep Dive until you ship something real.

Prova is structured to be honest. Not aggressive, not performatively harsh — honest. If a dimension scores 22, we say 22. If the bear case has a fatal flaw, we flag it before the bull case.

Founder

How Prova stays honest

The anti-sycophancy contract is wired into Prova's system prompt, then hardened at three layers of the stack:

• Personality: Prova is named, has four emotional states, never starts a message with "I", and never uses filler phrases like "great question" or "absolutely".
• Scoring: Every Deep Dive delivers a bull case and a bear case. If the bear case is empty, the dive is rejected at the schema layer.
• Commitment Gate: An application-layer block at commit_logic.py — not a warning, not a UX nudge. The next Deep Dive returns 403 with an explicit reason until commitments ship.

Tech stack

What we don't do

• · No generic advice. Every recommendation cites real data from the scan.
• · No flattery. If your idea scores 22, we say 22.
• · No idea data shared with other founders. Your inputs never train public models.
• · No silent score inflation. No pay-to-publish score boosts.
• · No new Deep Dive without completed commitments from the last dive.