Audit

Audit returns everything Aurra knows about how a memory came to exist: what input produced it, which LLM extracted it, when, and every event that affected it since.

Audit vs timeline. The timeline endpoint shows every version of a fact. Audit shows every event that happened to one specific memory. Use timeline to reconstruct state over time; use audit to answer "where did this fact come from and what happened to it".

GET `/memories/{memory_id}/audit`

Get the full audit record for a single memory.

Path parameters

Param	Type	Notes
`memory_id`	UUID	The memory to audit. Must belong to your account.

Example

curl https://api.aurra.us/memories/ba01899b-6294-4d53-8c04-2dd651e9500d/audit \
  -H "Authorization: Bearer $AURRA_API_KEY"

Response (200)

{
  "memory_id": "ba01899b-6294-4d53-8c04-2dd651e9500d",
  "memory": {
    "summary": "Alice is a senior engineer at Acme Corp.",
    "decision": "Alice is a senior engineer at Acme Corp.",
    "topic": "people"
  },
  "provenance": {
    "source_type": "agent",
    "channel": "agent",
    "tenant_id": "acme-demo",
    "workspace_id": null,
    "org_id": null,
    "external_id": null,
    "original_input": "Alice is a senior engineer at Acme Corp.",
    "captured_at": "2026-05-05T23:09:59.063114+00:00"
  },
  "extraction": {
    "model": "claude-opus-4-7",
    "prompt_version": "v1",
    "extracted_at": "2026-05-05T23:09:59.063114+00:00"
  },
  "temporal": {
    "valid_from": "2026-05-05T23:09:59.008804+00:00",
    "valid_to": null,
    "superseded_by": null,
    "is_current": true
  },
  "history": [
    {
      "event": "created",
      "at": "2026-05-05T23:09:59.063114+00:00",
      "by": "aurra-extractor",
      "details": null
    },
    {
      "event": "memory.created",
      "at": "2026-05-05T23:10:01.774058+00:00",
      "by": "user_3CyGtiPsbcMjBvQYnsU47r4hWKF",
      "details": {
        "shape": "legacy_content",
        "topic": "people",
        "source": "agent",
        "tenant_id": "acme-demo"
      }
    }
  ]
}

Top-level fields

Field	Type	Notes
`memory_id`	UUID	Echo of the requested ID.
`memory`	object	The current fact: `summary`, `decision`, `topic`.
`provenance`	object	Where this memory came from. See below.
`extraction`	object	Which LLM / prompt version generated it.
`temporal`	object	Current validity state. Same shape as in `/agent/memories`.
`history`	array	Event log in chronological order. See event types below.

provenance

Field	Notes
`source_type`	Top-level category (`agent`, `agent_session`, `slack`, `notion`, etc.)
`channel`	Sub-channel (e.g. Slack channel name, Notion database ID)
`tenant_id`	Tenant scope (if any)
`workspace_id`	Reserved for multi-workspace accounts (Phase 4).
`org_id`	Reserved for multi-org accounts (Phase 4).
`external_id`	Your system's identifier for the source event (e.g. Slack message ts).
`original_input`	The raw input that produced this memory. Trust floor.
`captured_at`	ISO 8601 timestamp of capture (not extraction).

extraction

Field	Notes
`model`	LLM identifier (e.g. `claude-opus-4-7`).
`prompt_version`	Aurra prompt version (e.g. `v1`). Bumps when Aurra changes extraction behavior.
`extracted_at`	ISO 8601 timestamp of extraction.

history[] events

Each entry has event, at, by, details. Common event types:

Event	Meaning
`created`	Extraction completed. `by` is always `aurra-extractor`.
`memory.created`	Memory row inserted into the database. `by` is the company's Clerk user ID.
`memory.superseded`	This memory was superseded. `details.superseded_by` has the new UUID.
`memory.expired`	This memory was superseded without a replacement. `details.valid_to` has the expiry.
`memory.classified`	Level 2 auto-supersession classifier ran against this memory. `details` has `verdict` (`supersedes`/`refines`/`independent`), `confidence`, `model`, `new_memory_id` (the new memory triggering the check), and `applied` (whether the verdict actually triggered supersession).

New event types may appear over time. Clients should accept unknown events without erroring.

Use cases

Debate resolution - "why does my agent think X?". provenance.original_input is always the answer.
Compliance reports - export provenance + extraction + history for any fact.
Extraction regression tracking - filter by extraction.prompt_version to quantify quality changes across extractor releases.

Error responses

Status	Trigger
`403`	`memory_id` exists but belongs to another account.
`404`	`memory_id` does not exist.
`422`	UUID format invalid.

On this page