Give AI a memory
of the work.
My ongoing experiment in treating ChatGPT as an interface to persistent operational state—so projects, decisions, tasks, and execution evidence survive a single conversation.
A great conversation shouldn't reset the project.
AI can help plan and reason in one session, then miss the current operational state in the next. If I have to re-explain commitments and decisions, the assistant hasn't become a reliable system.
Scattered context
Plans, corrections, tasks, and decisions were spread across chats and tools.
Continuity by default
Retrieve the latest project state and resume work with the next concrete action.
A persistent layer beneath the conversation.
Supabase stores state. ChatGPT supplies a conversational interface. Python routines and my personal dot, naDot, extend selected workflows. The diagram is interactive: select a component to see its role.
ChatGPT is the natural-language interface for retrieving project context and coordinating authorized work.
Real persistence, incomplete automation.
The operational database exists, with structured projects, tasks, context, activity and handoff records. Some cross-chat verification and synchronization mechanisms have been implemented; the overall system is still being developed and tested.
Persistent project state
Structured records make it possible to look up current work rather than reconstruct it from conversation history.
Cross-chat continuity
Synchronization, provenance, and checks are being refined. Not everything automatically enters the shared state.
Accountable execution
A trustworthy assistant that takes authorized actions, reports verifiable outcomes, and preserves the evidence.
Connecting software
to the factory floor.
Designing operational handoffs from order intake to fulfillment.