Auditing

Every tool call against your workspace is recorded, so you always have an answer to "what did my AI actually do?"

What's recorded

Each entry captures:

• the tool name and a redacted summary of its arguments,
• the result code and how long it took,
• the key that made the call and the user it acted as,
• the build id (for build-journey calls),
• the support id (Sentry trace),
• the client's user agent and a hashed IP address,
• the timestamp.

What's redacted

The audit log records the shape of a call, not its contents. It does not store your prompts, the field values you wrote, message bodies sent through send_via_integration, or enrichment inputs. Enough to answer "which records were updated by which key, when"; not enough to leak the data itself.

Retention

Audit entries are kept for 30 days, then deleted automatically.

Reading your log

Open MCP keys in the platform and choose View audit on any key to see its recent activity (no support request needed). This is the fastest way to confirm what a key has been doing and to spot a key that should be revoked.