BUILT THROUGH SPIRAL ONE
AI Memory Card
Your best ideas shouldn't disappear into old chats.
OVERVIEW
Your best ideas shouldn't disappear into old chats.
I use different AI models for different things. That worked really well until I actually wanted the knowledge to move with me.
AI Memory Card turns conversation exports from different AI providers into a searchable, portable archive that stays under the user's control.
Your AI usage should be transferable. Your history should be portable. And you should own it.
IMPLEMENTATION NOTES
Ingestion + normalization pipeline
AI Memory Card starts with user-provided exports rather than live account access: ChatGPT / Claude / Gemini Export → Provider Parser → Normalization → Deduplication → Structured Archive → Search / Recall.
Normalization preserves useful conversation metadata alongside actual content. Deduplication prevents repeated imports from quietly turning the archive into multiple copies of the same history.
A recorded import produced 29,453 entries from 327 conversations across four input files.
IMPLEMENTATION NOTES
Portable recall
Portable recall helps move relevant history into another environment as text rather than requiring the next model to have direct access to the original provider.
It isn't direct model-memory injection. It isn't live synchronization. It isn't vector search. It's a deliberately simpler portability mechanism.
The user has an artifact. The artifact is readable. It can move.
IMPLEMENTATION NOTES
Local default + optional webhook
The default workflow is local. Exports are supplied by the user, processed locally, and turned into local archive and recall artifacts.
There is an optional webhook path for deliberately configured workflows that need to send structured data elsewhere.
AI Memory Card is an archive and portability utility. It does not provide live provider synchronization, semantic or vector search, direct memory writing, guaranteed perfect extraction or guaranteed improvement in future model responses.
BOUNDARIES
What this is not claiming.
The evidence viewer keeps the limitation attached to each artifact. Current capability is shown separately from future direction so the technical story stays useful without getting ahead of the work.
DIRECTION
What I am still testing.
- Reviewed public demonstration
- Reviewed public sample workbook
- Current Windows verification