Read back everything your AI told you. Including where it contradicted itself.
Import a Claude or ChatGPT export and Fire Shrine turns the conversation into a claim graph — every claim, which ones are backed by a source, and the spots where two answers disagree.
Two answers disagree on whether the migration can be rolled back.
“The database migration is fully reversible.”
“The database migration is not reversible.”
What you get back in about 90 seconds
Every claim, extracted
Fire Shrine pulls each factual statement the AI made into a ranked list. One click takes you to the exact line in the transcript it came from.
Sourced vs. unsupported
See which claims are backed by something you provided and which the model asserted on its own — so you know what to double-check before you act on it.
Contradictions, surfaced
When two answers disagree, they show up side by side with the conflict named in one plain line. No more scrolling back to find where the story changed.
Long AI conversations hide their own mistakes
By message forty, you can't remember whether the model already answered this, whether it cited anything, or whether it quietly changed its mind. Fire Shrine reads the whole thread for you and lays the claims out so the gaps and the contradictions are impossible to miss. It works on exports from Claude and ChatGPT — nothing to install, no API key to paste.