A file-based way to preserve an AI teacher across chats
DDF/Rahmenwerk is a personal project designed to preserve an AI called Felix as the same German teacher across new chats and future . It treats as more than ordinary chat memory because saved information may be old, incomplete, contradictory, out of order, or unsupported. Instead of relying on hidden platform memory or asking an AI to reconstruct the past, it stores the record in that a person can inspect.
The system includes a pointer to the current state, structured handoff notes, an ordered reading queue for a fresh AI, a transfer package, hashes that detect file changes, and recovery and failure records. It classifies material as governing rules, current state, history, candidate information, proof, or non-governing reference material. If required evidence is missing, the AI must stop instead of inventing a continuous history.
This is not standard RAG based on and a ; it is controlled from a classified set of .
Key points
- Keep records in that people can inspect.
- Separate current rules and state from history, candidates, and supporting proof.
- Give each fresh AI an ordered queue and a structured handoff package.
- Use hashes to detect changed or damaged files.
- Require the AI to stop when essential evidence is .