The memory layer for AI-assisted ML development
Segue keeps your model project's goal, evaluation rules, best known model, decisions, failed approaches and next steps in one brief your assistants can load. Keep Git for code and your experiment tracker for runs.
Works over MCP with OAuth · ChatGPT · Claude · Claude Code · Cursor · Codex
Across tools and sessions
One project brief. A useful starting point in every AI.
Start with the nine-section ML template and the material you choose to save.
Load the same handle. Get a compact snapshot, then open the sections you need.
Review the complete proposed changes in Segue before you approve them.
Built for work that changes
Keep the learning, the evidence and the next decision connected.
No gain on the held-out set.
Record the outcome and the reason.
Keep the outcome and the reason, so your next session can build on what you learned.
Link the commits, runs and datasets behind a decision. Your artifacts stay in their original tools.
Check this conclusion against the new data.
Spot stale answers and changed inputs before you carry a conclusion into the next session.
The earlier value becomes current.
Still in your section history.
See earlier section values. Restore one as a new revision while keeping the history.
Segue does not record conversations in the background or run model inference on your content. Your assistant reads the material you select. Export and delete your memory on any plan.
No. Use plain Git, or link your tracker runs from the brief.
Yes. Keep repository conventions there and link the evolving project state in Segue.
Segue