- persist key facts extracted from documents (parties, dates, issues)
- give agents per-case context that survives across sessions
- build a correction layer so agents learn from feedback
Entry types
Every memory entry has atype that describes the kind of knowledge it holds:
| Type | Use case |
|---|---|
fact | General statements (“Client prefers email over phone”) |
party | People and organizations involved |
issue | Legal or business issues identified |
deadline | Important dates and filing deadlines |
discovery | Discovery-related findings |
correction | Corrections to previous agent outputs |
preference | User or client preferences |
Create an entry
Endpoint
Request body
| Field | Type | Required | Description |
|---|---|---|---|
type | string | Yes | One of: fact, party, issue, deadline, discovery, correction, preference |
content | string | Yes | The memory content |
source | string | No | Where this fact came from (e.g. intake-form, agent-extraction) |
tags | string[] | No | Tags for filtering |
Response
Search memory
Search entries by meaning — not just keywords.Endpoint
Parameters
| Field | Type | Required | Description |
|---|---|---|---|
query | string | Yes | Natural-language search query |
types | string[] | No | Filter by entry type |
tags | string[] | No | Filter by tag |
limit | number | No | Max results (1–100, default 10) |
List entries
Retrieve all memory entries for a vault.Endpoint
Update an entry
Endpoint
Parameters
| Field | Type | Required | Description |
|---|---|---|---|
content | string | No | Updated content |
source | string | null | No | Updated source (set null to clear) |
tags | string[] | No | Replacement tag list |
Delete an entry
Endpoint
{ "deleted": true } on success.
