Chorus
Is that answer repeatable?
An AI's answer hands you one confidence level for the whole thing. Chorus redistributes that confidence claim by claim: it re-asks your exact conversation several times, has a judge compare every substantive claim across the runs, and tells you — with evidence — what the model stands behind, what it won't back on a re-ask, and what to change to earn the confidence you thought you had.
How to use it
- Create an account — Chorus is invite-only right now; you'll need an invite code.
- In Claude: Settings → Connectors → Add custom connector, and paste:
https://chorus.ligament.me/mcp - Sign in when Claude asks — once.
- In any conversation: “Ask Chorus: is this repeatable?”
What you get back
A verdict — repeatable or not — with every disagreement between runs quoted, and, when a check fails, the Close: what held steady, what your answer overclaims, which missing facts would pin the wobble down, and what only a source outside the model can settle.
Your data. When you run a check, your conversation passes through Chorus's server
and the Anthropic API. Raw transcripts are not stored: after a check completes, Chorus retains only
the derived results — the claim table, verdict, and analysis. Chorus measures repeatability, not
truth: a consistent answer is not a certified fact.