Agentic FabriqOne trail across everything.
Every agent, every user, every connected tool — a single centralized record, one schema.
PLATFORM / Audit Trail
One audit trail. Every agent, every user, every tool.
All agent activity across all your applications lands in one centralized record — both identities, the action, and the verdict.
01 / IDENTITY02 / POLICY03 / AUDIT
01 / THE CONTEXTA record the agent
A record the agent
does not write itself.
When an agent misbehaves — or an auditor asks — you need a record you can trust. The agent shouldn’t write its own.
AUTHORITY
agent
∩
person
RESOLVED PER REQUEST∩
person
02 / IN PRACTICE
Every action leaves a thread.
THREE RECORDS · ONE TRAILIllustrative · the same schema whatever the verdict
EVT 8f2a·0431
Google DriveRead an authorized record- asked by
- Dana Whitfield · Analyst
- acting agent
- review-agent
- call
- drive.read · Q3 board pack
- policy
- Within both grants
- identity
- scope
- policy
Both identities on the record, and the reason it was allowed sits next to them.
EVT 8f2a·0432
SalesforceDelete a protected record- asked by
- Priya Menon · Support
- acting agent
- support-bot
- call
- crm.record.delete · account 4471
- policy
- User has no delete scope
- identity
- scope
- policy
A refusal is an event too. The trail says which check stopped it, not merely that nothing happened.
EVT 8f2a·0433
BigQueryExport an audit report- asked by
- Dana Whitfield · Analyst
- acting agent
- review-agent
- call
- audit.export · last 30 days
- policy
- Awaiting a named approver
- identity
- scope
- policy
The wait is on the record from the moment it starts — not only once someone answers.
Illustrative records · the agent does not write these; it is written about.
03 / WHAT CHANGES
Control, in the details.
Both identities on every event.
The person who asked and the agent that acted — with tool, action, and timestamp.
The verdict is part of the record.
Allowed, held for approval, or refused — outcomes land in the trail.
Query and filter in the dashboard.
Slice by service, event type, and time window.
Warehouse-grade storage, exportable.
Events land in BigQuery-backed storage and export to your own tooling.
Usage metering alongside.
Calls per user and per tool, over time — right beside the audit trail.
THE INDEPENDENT RECORD
Follow the action backward.
Illustrative activity ledger
10:42:08drive.readresearch-agent for Danaallowed
Tool: Google Drive · Agent: research-agent · Acting user: Dana · Policy: Within both grants
Request→Identity resolved→Policy checked→Decision recorded
10:42:04drive.deletesupport-bot for Priyablocked
Tool: Google Drive · Agent: support-bot · Acting user: Priya · Policy: User lacks delete scope
Request→Identity resolved→Policy checked→Decision recorded
10:41:59
github.open_prcoding-agent for Alexallowed
Tool: GitHub · Agent: coding-agent · Acting user: Alex · Policy: Branch write permitted
Request→Identity resolved→Policy checked→Decision recorded
10:41:36
slack.post_messagesupport-bot for Priyaallowed
Tool: Slack · Agent: support-bot · Acting user: Priya · Policy: Channel within both grants
Request→Identity resolved→Policy checked→Decision recorded
10:41:12salesforce.exportresearch-agent for Danablocked
Tool: Salesforce · Agent: research-agent · Acting user: Dana · Policy: Export not in the agent’s scope
Request→Identity resolved→Policy checked→Decision recorded
THREE LAYERS. ONE DECISION.
The boundary travels with the work.
Explore the context attached to every request.
What is this agent allowed to do?
review-agent has a defined purpose and a bounded set of tools.
Whose authority travels with it?
The request acts for analyst, using that person’s permissions.
Where does the action land?
The policy applies to the requested action within Activity ledger, before the tool executes.
QUESTIONS
A closer look.
Is it really one trail across everything?+
Yes. Every agent action against every connected tool lands in the same record — one schema for all users and agents, no per-app log stitching.
Does the agent write its own log?+
No. Logging happens at the policy layer where each call is checked — outside the agent’s reach entirely.
Are refused calls in the trail too?+
Verdicts are part of the record — events show whether a call was allowed, held for approval, or refused, right next to who asked and which agent acted.
Can we export to our own tooling?+
Yes — events live in warehouse-grade storage and export cleanly for your SIEM, warehouse, or audit workflow.
How do we investigate an incident?+
Filter the trail by service, event type, or time window in the dashboard — every event carries both identities and the verdict, so the sequence reconstructs in minutes.