Skip to content

Audit Log & Compliance ​

The audit log records every significant action taken in your DMOps.ai tenant. Every entry is signed and immutable.


What is recorded ​

Event typeExamples
Task eventsTask created, status changed, assigned, deleted
Agent runsAgent picked up a task, completed, produced an artefact, blocked
Tool callsAgent called a tool, with tool name, arguments, and result summary
Approval eventsApproval requested, approved, rejected, timed out
Admin changesAgent definition changed, integration added, user role changed
Auth eventsLogin, logout, API key created/revoked
Cost eventsToken usage recorded per task and per agent

Viewing the audit log ​

Go to Observability → Audit Log.

You can filter by:

  • Date range
  • Actor — human user or agent name
  • Event type — filter to a specific category
  • Task ID — see all events for a specific task
  • Keyword — full-text search across log entries

Execution traces ​

For agent task runs, DMOps.ai records a full execution trace — the step-by-step actions the agent took:

  • Each LLM call (prompt, model, token count, cost)
  • Each tool call (tool name, arguments, response)
  • Each artefact created
  • Time elapsed

To view a trace:

  1. Open a task from the task board.
  2. Click the Trace tab in the task detail panel.
  3. Expand any step to see the full prompt or tool payload.

Exporting audit data ​

Coming soon

CSV and JSON export of the audit log for compliance reporting is on the roadmap.


Retention policy ​

By default, audit logs are retained for 90 days. For compliance use cases requiring longer retention, configure an OTLP sink or Webhook sink in your integrations to stream events to your own data store.

See Connecting Integrations → OTLP Sink / Webhook Sink for setup instructions.


Compliance notes ​

  • All credentials are encrypted at rest (AES-256) with your tenant's master key
  • Task data and agent outputs are stored in your dedicated database (no shared data between tenants)
  • API traffic is TLS 1.2+
  • Agents cannot self-approve; every approval gate requires a human actor
  • Token costs and tool calls are logged per-task for financial and operational accountability