Dashboard
Activity log and RevealNew
Review every checked prompt, its masked values and the AI's answer, and reveal encrypted originals with an audited action.
Open an AI system and select Activity (Logs). Each row is one prompt or event; the guard check and the logged answer of the same model call are joined into one row.
Row status
| Badge | Meaning |
|---|---|
| Prompt checked | Guard checked the prompt; no answer has been logged (for example a guard-only integration) |
| Awaiting output | The prompt was sent and AgentID is waiting for the answer to be logged |
| Complete | The answer was logged |
| Output not captured | The integration could not capture the answer |
| Security Alert | A detector or policy flagged the prompt without blocking it |
| Security Block | The prompt was blocked before the model call |
Shadow Mode rows show the simulated outcome; use Actual vs Simulated to separate them.
Filters and export
- User: search by e-mail, name or user ID.
- Event Type, Actual vs Simulated, From and To dates. Reset clears the filters.
- Export as CSV or Export as JSON for the current view.
Log detail
Click a row to open its detail. Depending on the integration you see:
- the prompt and the answer as stored (masked text, or
[REDACTED]when encrypted storage is on), - detected personal information types and security signals, with the detection source,
- request identity, API key name, prefix, owner label and environment,
- attachment and paste checks, and for browser turns each response attempt (regenerations keep their own output),
- the raw event metadata.
Export JSON downloads the single event. Mark as False Positive records feedback on a detection.
Masked values
For browser extension turns the detail shows a Masked values table:
| Column | Content |
|---|---|
| Token | The placeholder, for example <PERSON_NAME_1> |
| Sent to AI provider | What the provider actually received for that value |
| Original | [ENCRYPTED] until revealed, or Not stored |
The originals are stored only when Store Encrypted Payloads is enabled in Guardrails. Without it, the table says that originals are not stored.
Reveal encrypted content
When a row has encrypted content, admins of the AI system see Reveal payload in the detail header.
Open the row
Open the event in Activity (Logs).
Click Reveal payload
AgentID decrypts the prompt, the output and the originals behind each placeholder and shows them in the detail. Hide payload hides them again.
Review the audit trail
Every Reveal is written to the audit log as ENCRYPTED_EVENT_PAYLOAD_VIEWED, with which parts were revealed. Organization admins can review it in the Global Audit Log.
Workflows
Agent runs that log tool calls and steps with a shared workflowRunId appear as a workflow summary row that opens a step-by-step timeline. The guarded prompt of each LLM step stays inspectable as its own row. Non-model steps show Model: Not applicable and do not count toward spend.