Memory
Brody AI keeps cross-session memory of your project so it never feels amnesiac — it recalls what it learned last session, in this project, for you.
memory_sync
After meaningful project changes, the agent persists the new state to the memory store. Architecture decisions, tool outcomes, and scene facts get saved so future sessions start from the current reality — not a cold read.
memory_query
When a prompt depends on prior work, the agent queries memory with relevant keywords before planning. The persona rules require it: if the user refers to earlier sessions, look it up first.
Memory APIs
| Method | Endpoint | Purpose |
|---|---|---|
| POST | /api/v1/memory/search | Semantic search over stored project memory for the current user + project scope. |
| POST | /api/v1/memory/prune | Remove stale or unwanted memory entries from the store. |
Base gateway URL: NEXT_PUBLIC_GATEWAY_URL (defaults to http://localhost:4000 in dev) — same host as the chat API (/api/v1/chat).
Privacy & scoping
- Memory is scoped per user and per project — your memory is never visible to other accounts or other projects.
- Cross-session means your sessions, not shared sessions.
- You can prune stored entries at any time via the
/api/v1/memory/pruneendpoint.
Keeping memory tidy
Stale entries (from discarded experiments or renamed systems) can confuse later sessions. Pruning the memory store removes them and keeps query results relevant — the agent also skips memory when a prompt does not depend on prior state, so it never drags old context into fresh work.
Related: tools · verification