Honest AI, fairly priced

Each confessional is generated directly from the chat log you paste. The AI rereads your real transcript and writes its own first-person account of where it drifted, what it assumed, and what it would do differently. The model has nothing in its context except your session, so it can't fall back on a generic template. Paste a hallucination and you'll get a different confessional than someone who pasted a sycophancy loop.

You paste a chat log or describe what happened, and Confess runs a multi-step prompt: first it identifies the failure mode (hallucination, drift, fabricated detail, sycophancy, and so on), then it drafts the AI's first-person account, then it self-reviews for vagueness and rewrites the weak sections. A typical generation takes 20–40 seconds, and you can read, share, or download the result.

Your input is sent to the language model for this single generation and is not used to train any model. The generated confessional can be made public in the gallery at your discretion, or kept private. We don't store the raw input after the confessional is written — only the final document.