Memories
Memories are persistent facts about end-users that are automatically injected into the teammate's context during runs. Use them to personalize teammate behavior.
How Memories Work
- You create memories via the API (or the teammate saves them during runs)
- When a run executes for that
user_id, memories are included in the system prompt - The teammate uses these facts to personalize its responses and actions
User Metadata
Pre-populate memories via the API to give teammates immediate context about your users. Think of these as user profile data that personalizes every interaction.
Create a Memory
Each memory is a short text (max 300 characters). Duplicates are automatically detected (case-insensitive, whitespace-normalized) and return a 409 Conflict response.
List Memories
The source field is "api" for memories you create, or "agent" for memories the teammate saves during runs.
Delete a Memory
The user_id parameter is required on delete to prevent accidental deletion of another user's memories.
Pre-Populate During Onboarding
Seed memories before the first run to give the teammate immediate context:
Saved Memories (Built-in Tool)
During runs, the teammate can save memories via its built-in memory tool. These are stored with source: "agent" and follow the same isolation rules.
The teammate decides what to remember based on the conversation — things like user preferences, key decisions, or recurring patterns.
To disable saved memories for a specific run:
Previous Runs (Built-in Tool)
Runs automatically include context from previous runs for the same user. This lets teammates build on past work without you managing conversation history manually.
The teammate can search through previous run outputs to recall what was done before — giving it continuity across sessions.
To disable previous run history for a specific run:
Limits
- Max length: 300 characters per memory
- Deduplication: identical content (case-insensitive, whitespace-normalized) for the same user returns
409 Conflict - Scoping: memories are strictly isolated per
user_id
What's Next
- Users — multi-tenancy and data isolation with
user_id - Permissions — control tool access with approval modes