API referenceChat and agents

Replay Records

Look back at how a run came together: replay stores prompts, responses, and tool inputs and results. Durable chat runs write a record automatically under their runId; client-side agents can upload their own. Only the authenticated account can read its records through these endpoints.

Review before sharing. Credential filtering is best-effort, not anonymization. redacted: true means the filter ran, not that every sensitive detail was removed. Keep secrets out of prompts and tool data, and review any record you export or share.

#Store a record

POST /v1/replay/records Auth required

Stores a RunRecord and returns 201. The owner is derived from the authenticated wallet, so clients cannot write into someone else's namespace. Posting the same run_id again replaces the stored record. Malformed JSON or a missing required field returns 400.

#Body

Name	Type	In	Description
schemaVersionrequired	integer	body	RunRecord schema version. Currently `1` or `2`.
run_idrequired	string	body	Caller-chosen run id (max 128 chars). Uniqueness is per owner.
user_requestrequired	string	body	The user prompt or request that triggered the run.
roundsrequired	array	body	Recorded model messages, tool calls, and tool results.

Payload cap. JSON is capped at 1,000,000 UTF-8 bytes (1 MB), not characters. Larger payloads return 413.

#Response

{
  "runId": "run_…",
  "schemaVersion": 2,
  "redacted": true,
  "createTime": 1731950400,
  "updateTime": 1731950400
}

#List records

GET /v1/replay/records Auth required

Lists recent RunRecords for the caller, newest first. Summaries include the first 1,024 Unicode characters of each prompt and final response. Fetch an individual record for the full text and tool data.

#Query parameters

Name	Type	In	Description
limit	integer	query	Default `50`, clamped to 1–200. A non-numeric value returns `400`.
cursor	string	query	Omit for the first page. Pass the previous response's `nextCursor` unchanged to fetch the next page; a modified cursor returns `400`. Stop when it is `null`.

#Response

{
  "records": [
    {
      "runId": "run_…",
      "schemaVersion": 2,
      "createTime": 1731950400,
      "updateTime": 1731950400,
      "userRequest": "Storyboard a 5-shot neon teaser.",
      "finalResponse": "Here's your storyboard.",
      "modelId": "qwen3.6-35b-a3b-gguf-iq4xs",
      "rounds": 3
    }
  ],
  "nextCursor": null
}

#Read a record

GET /v1/replay/records/:id Auth required

Returns a full RunRecord, suitable for the replay viewer. Unknown ids return 404. The credential filter is applied again when records are read.

#Response

{
  "record": { "schemaVersion": 2, "run_id": "run_…", "rounds": [ "…" ] },
  "createTime": 1731950400
}