Lab / Concept
Multi-model reasoning
Delphieiux: ask once, a council answers
Several AI models answer the same question on their own, review each other blind, argue the other side, and a chair writes up what they agree on and what is still in dispute.
The idea
One AI model gives you one confident answer. Delphieiux asks several models the same question. Each answers on its own, then they review each other without knowing who wrote what, argue the other side, get scored, and a chair writes up what they agree on and what is still disputed, as a memo, a table or JSON that matches your schema.
Paying for the thinking
- Free: free models only, capped so it can never cost the operator money.
- Your own key: you pay your provider directly; the key is entered in the app and stored encrypted.
- Credits: planned, switched off for now.
Sign-in is by passkey, with role-based access for teams.
Status
In development. We want it free to use as far as we can make it, and it will be open source once it is complete. Until then the code is private.
More from the lab