Privacy & Portability
How to move project context from one AI to another
Short answer
Do not transfer only the conversations. Build a context package containing current state, decisions with their reasons, constraints, open questions and observed outcomes. That lets another AI resume the work without inheriting every piece of historical noise.

Switching AI tools feels easy until you realize that six months of project context stayed behind in the previous product. Exporting conversations is a useful first step. It is not yet a context migration.
A transcript tells you what was said. Project context tells you what matters now and why.
Start with current state
Before moving anything, create a dated summary containing:
- project objective;
- current state;
- active decisions;
- constraints;
- people or systems involved;
- open questions;
- next steps.
A new human collaborator should be able to understand the project from this document in ten minutes. That is a useful quality test for an AI too.
Add history only when it explains the present
Many old decisions are no longer useful. Some explain exactly why the project looks the way it does today.
For example:
Option A was tested and abandoned on June 3 because its licensing cost was incompatible with the business model. Do not propose it again unless that economic constraint changes.
That one sentence prevents the next AI from rediscovering the same dead end.
Separate data from instructions
An export may contain sentences that look like commands but are merely traces from an old conversation. Historical text should not automatically become a current instruction.
During migration:
- identify rules that are still active;
- mark superseded decisions;
- separate quotes and external data from instructions;
- preserve provenance;
- do not automatically promote an AI-generated summary to verified fact.
Portability of meaning is harder than portability of files
Downloading JSON or Markdown is valuable because you recover your data. The receiving system still needs to understand what each item means: decision, preference, fact, outcome, hypothesis, rule or historical note.
A technically complete export can therefore remain semantically weak.
A strong migration preserves as much as possible:
- information type;
- date;
- source;
- current status;
- relationship to replacement information;
- privacy level.
Run a restart test
After migration, ask the new AI five questions that the old one should have been able to answer:
- What is the current objective?
- Which important decision changed recently?
- Which option did we already reject and why?
- What is still uncertain?
- What should happen next?
If those answers are accurate, you have transferred more than history.
Do not forget privacy
Moving everything into a new tool does not mean the new tool should automatically receive every personal or private space. Portable context still needs boundaries.
The better question is not only, “Can I export?”
It is, “Can I carry the useful context, with its meaning and its boundaries?”
The confusion between data and instructions is detailed in importing memory without importing hallucinations and why found does not mean true. On portability of meaning, see data portability vs portability of meaning and what a real AI memory export should contain. To connect the other AI afterwards, the tutorial connect ChatGPT to Nawame shows the procedure. The question of boundaries is picked up in sovereignty and AI memory, on the security page and in the documentation.
Limits
There is no universal standard today that guarantees one provider’s memory export will be interpreted in exactly the same way by another. Formats, memory models and privacy controls continue to evolve. Important migrations still deserve human verification.


