How do I move AI memory between projects?

Export or ingest the relevant conversations and notes into a user-owned memory vault, convert them into canonical records, index them, and retrieve only the project-specific context needed for the next coding or writing task.

Does moving AI memory train the model?

No. Adamant is a retrieval workflow. It stores and searches user-owned context, then returns cited snippets to a model when the user asks for them.

Back to Adamant

Move AI memory between projects without losing context

If a model forgets the decisions from your last repo, workspace, or project, the problem is not your prompt. The problem is that memory is trapped inside a tool-specific silo.

Short answer: Adamant is designed to turn conversations, notes, and project history into a portable memory vault. A coding agent such as Codex or Claude Code can query that vault for a small cited context bundle instead of asking you to paste the same background over and over.

Common symptoms

You start a new repo and the AI no longer knows the architecture decisions from the old repo.
A Claude Project has useful context, but Codex or Cursor cannot see it.
Your ChatGPT conversation explains the product, but Claude Code starts from zero.
Important debugging history is buried in a chat thread that is not connected to the current task.

The portable memory workflow

1. Import

Bring in ChatGPT exports, Claude notes, Markdown docs, project notes, and future connector records.

2. Index

Chunk, sanitize, tag, and embed the records into local or user-owned vector storage.

3. Retrieve

Ask Adamant for the exact context needed by the current project, model, or tool.

Example coding-agent query

python -m adamant context "what prior decisions affect this auth refactor?" \
  --mode local \
  --top 5 \
  --max-tokens 1200

Why this is different from model memory

Platform memory usually stays inside one provider. Adamant keeps the memory layer outside the model. That lets the same source context support ChatGPT, Claude, Gemini, Codex, Claude Code, Cursor, local models, and future tools.

What the model receives

The model does not receive your whole vault. It receives a narrow, cited bundle with titles, source IDs, scores, and relevant snippets. That keeps project context useful without flooding the model or copying unnecessary personal data.