Agent memory

Agent memory

ParseRail stores, searches, and forgets memories for your agents in your own namespace. It provides semantic recall without a vector DB to run.

Cost: 2 credits ($0.02) — charged only on success.

post/v1/memory

Request

  • Base URL: https://api.thecompound.tech
  • URL: https://api.thecompound.tech/v1/memory
  • Auth: HTTP bearer

Headers

ParseRail-Dry-Run'1' | 'true'

Send 1 to dry-run the call. ParseRail runs auth, the throttle, body validation and the balance precheck. It returns 200 with the cost and the status a live call would get. No model runs and nothing is charged, so a key with a zero balance works. Dry runs count against the per-key rate limit.

Idempotency-Keystring

Key for one logical operation. Same body replays the result for 24 hours; a different body with the same key returns 409.

Request body

op'store' | 'search' | 'forget' required

The operation.

namespacestring

Provide your partition key, such as an agent id or a user id. It defaults to "default".

contentstring

store: the memory text.

metadataobject

For store:, provide attached metadata; recall returns it.

querystring

search: what to recall.

limitinteger

search: max results.

idstring

For forget:, provide the memory id, or omit it with namespace to wipe it.

Response

The call succeeded. The response contains result fields and a usage envelope. With ParseRail-Dry-Run: 1 the body is a DryRun object instead.

OR

Changes

No recorded changes to this operation across all 1 revision of this API.