Memory
What Alvin remembers about you, how to see it, and how to control it.
Memory is what lets a conversation start where the last one ended. Alvin picks up durable facts as you work — what you're building, how you like answers, constraints you've stated — and carries them into later sessions and across agents.
What gets remembered
Memories are short factual statements, each in a category:
| Category | What it holds |
|---|---|
| Identity | Who you are — role, background |
| Preference | How you like to work |
| Goal | What you're working toward |
| Project | Context specific to a project |
| Relationship | People, teams and systems you work with |
| Constraint | Hard limits or rules you've stated |
Passing detail from a single conversation isn't kept. The intent is durable context, not a transcript.
Personal and project memory
Personal memory follows you across every conversation and every agent.
Project memory is shared. Memories contributed in a project's context are visible to everyone with access to that project, so a team builds shared context rather than each person re-explaining the same thing.
Seeing and controlling it
Memory is inspectable. In Studio you can review what's been recorded, edit anything that's wrong, and delete anything you'd rather wasn't kept. Deleting a memory removes it from future conversations.
If an answer seems to assume something outdated about you, check your memory — an old preference is the usual cause, and deleting it fixes it immediately.
Memory and Genie
Genie draws on the same memory. That's much of what makes it feel continuous rather than starting over each session.
Programmatic access
Memory is readable and writable through the API — list, add, update, delete, and incrementally sync. See the Memory API.
Related
- Genie — the persistent personal agent
- Projects — where shared memory lives
- Memory API — programmatic access