Memory
Remember a fact
Store one fact so the agent can recall it later by meaning. The text is embedded on write; retrieval is semantic, not keyword.
Scope goes in the body here. Pass scope plus the matching contactId / conversationId. Omit it for the project scope. Every read endpoint takes the scope as a ?scope= query parameter instead — see the scope parameter on GET /v1/memory.
One fact per call. A paragraph carrying five unrelated facts is embedded as an average of all five and retrieved well by none of them.
POST
Authorizations
Bearer authentication header of the form Bearer <token>, where <token> is your auth token.
Body
application/json
Store one fact. It is embedded on write, so it becomes searchable by meaning rather than by keyword.
What to remember. Max 8,192 bytes of UTF-8. Write one fact per call: a paragraph holding five unrelated facts embeds as an average of all five and is retrieved well by none of them.
Example:
"Prefers to be contacted in the morning, and by WhatsApp rather than email."
Which namespace a memory belongs to. Scopes are isolated: a search in one never returns a memory from another.
project: shared by the whole project. What the agent knows in general.contact: what the agent knows about one person, across every conversation with them.conversation: what the agent knows about one inbox thread.
Defaults to project everywhere. contact and conversation require the matching id.
Available options:
project, contact, conversation Required when scope is contact.
Required when scope is conversation.
Arbitrary string tags stored alongside the fact and returned with it. Keys max 64 chars, values max 512, whole object max 2,048 bytes. Not searchable — search matches text only.
Delete the fact after this many seconds. 60 to 63,072,000 (2 years). Omit to keep it until deleted.
Required range:
60 <= x <= 63072000Response
Fact stored.
A stored fact, as returned when it is created.