MY MEMORY
Your memory should belong to you.
A research and working infrastructure for continuity across long-running AI-assisted work.
Loss of continuity
The question
What remains of a long-running human project when the conversation ends, the model changes, or the system that helped with the work is no longer there?
A serious project can outlive a chat, a session, a model and eventually an AI provider. When that happens, decisions become separated from their reasons, sources disappear from conclusions, commitments drift out of view, and stale information begins to compete with what is actually current.
The problem is not simply that an AI “forgets”. The deeper problem is that the history of the work — what happened, why a decision was made, what later changed and which state is now authoritative — can become fragmented across systems that neither the person nor the next AI fully controls.
What should remain
MY MEMORY began from a practical need: a successor AI should not have to guess what happened before it arrived, and the person should not have to reconstruct the same project history again and again.
Project → Event → Decision → Reason → Person → Document → Commitment → Current State → Source
The AI can participate in memory. It should not own the memory.
The working system therefore keeps durable project memory outside any single conversation or provider and treats provenance as part of the meaning. A fact is not the same thing as an assumption. A proposal is not the same thing as an accepted decision. Retained history is not automatically current truth. A later correction should be able to supersede an earlier state without pretending that the earlier state never existed.
Why user-owned continuity matters
Continuity is not only about retrieving information. It is about preserving the structure of how understanding was reached.
If decisions, reasons, sources and context are distributed across systems that can change or disappear, then a person can retain the files and still lose the history that made those files meaningful.
MY MEMORY explores a different relationship with AI: the model may help capture, connect, retrieve and explain memory, while the durable record remains controlled by the user and can survive the replacement of the model itself.
The system may change. The continuity should remain with the person.
System
How it works
The current architecture is a provenance-aware continuity system for maintaining explicit project state across sessions, models and AI providers.
Its implemented mechanics distinguish current, stale, superseded, conflicting and uncertain information; preserve decisions together with reasons and sources; and use explicit human-reviewed state transitions rather than treating opaque conversation retrieval as authoritative memory.
Recovery work has exercised clean successor sessions, adversarial conditions, stale-state rejection and cross-provider recovery. The aim is not to make every model reproduce the same answer. It is to make the state of the work recoverable enough that a new system can understand what is known, what is uncertain, what was superseded and what still requires a human decision.
Evidence
Evidence so far
Clean successor recovery
In one preregistered clean successor-session run, a fresh ChatGPT session recovered the specified state of a real long-running project from user-owned Google Drive artifacts.
Project-history retelling
The author did not retell the project history, no facilitator help was used, and the successor did not request a project-history recap.
Portable recovery tests
Bounded recovery work has also exercised multiple AI providers, adversarial successor conditions, stale-basis rejection and explicit human acceptance of a fresh return.
External-user workflows
Two independent external-user runs completed with zero project-history retelling and zero facilitator help. This is directional evidence only, not a general usability claim.
What this evidence does not establish
- A finished general-purpose memory system.
- Production security, scale or multi-user readiness.
- Provider-neutral writable interoperability across all AI systems.
- Long-duration reliability under production conditions.
- Any clinical, medical or cognitive-health benefit.
The current evidence supports bounded continuity and recovery behaviour under the conditions tested. Claims should not be extended beyond those conditions.
Now
What is being explored now
The present work focuses on Project & Decision Memory for real, multi-stage AI-assisted projects: preserving decisions, reasons, sources, commitments and current state across many sessions and tools while reducing reconstruction work and stale-history errors.
The next practical questions are how to make meaningful capture less manual, how to retrieve only the context that actually matters, how to keep conflicting or uncertain records visible without collapsing them into false certainty, and how to make continuity useful in ordinary daily work rather than only in controlled recovery tests.
Current state
- Working prototype.
- Used in ongoing long-running AI-assisted project work.
- Bounded continuity and recovery tests completed.
- External-user workflows have been tested on a small scale.
- The architecture remains under active development and review.
Human
Long-term direction
The larger idea is personal continuity: memory that remains with the person as years pass, projects change and AI systems come and go.
The long-term aim is not merely to make one AI conversation remember another. Continuity between sessions, models and providers is an enabling mechanism. The deeper question is whether a person can preserve and later recover the structure of their own history: what happened, what they knew at the time, what they decided, why they decided it, what sources shaped that decision, what changed later and which version of the record is current.
The same continuity principles may eventually extend into organizational, research, creative and other long-lived knowledge environments where provenance and current state matter.
Future research. A separate horizon is personal cognitive continuity over long periods of life, including ordinary human forgetting and, separately, whether user-owned memory structures could one day be useful in research involving early memory decline or early-stage dementia. This is not a medical claim. MY MEMORY is not a medical device, diagnostic system, dementia treatment or clinical intervention; any such application would require dedicated research, safety work and appropriate validation.
Origin
MY MEMORY grew from the practical need to preserve continuity across Ashraellen's own long-running AI-assisted projects, where decisions, reasons, sources and current state had to survive beyond individual sessions and tools.
I did not begin building this system as an abstract experiment in AI memory. While working across many parallel projects, it became increasingly difficult to keep in mind where something had happened, what had already been decided and why. Part of that continuity therefore had to move outside my head — into project state, decisions, sources, handoff and recovery.
What began as a solution to repeated context loss gradually became a broader question about memory, authorship, control and continuity in a world where human work increasingly passes through changing AI systems.
At first I needed memory for projects. Then the question became larger: what should a person be allowed not to remember alone?
The system may change. The memory should remain with the person.
Contact
Research dialogue
For research discussion, testing or relevant collaboration: