Product · Supervision

Supervision-grade evidence for the AI tools your employees already use.

Your team already works in Claude Enterprise or ChatGPT Enterprise. When a regulator, auditor, or client asks how that use is supervised, a provider dashboard is not an answer. Tallin captures the activity through the provider's compliance API, archives it message by message in tamper-evident storage you set the terms on, and gives your compliance team a working review queue: reason-gated access to flagged content, dispositions from your own policy, and an audit trail that shows every decision.

Request access →

How capture works

Captured from the source of record, archived before it expires.

Tallin reads sanctioned AI activity from the provider's own compliance API, the same interface the provider offers for regulated customers, and lands every message in an append-only archive with a hash chain per conversation. The polling cadence is designed to beat the provider's retention window, so your record outlives theirs.

Claude Enterprise and ChatGPT Enterprise connectors, reading through each provider's compliance API.

Message-level capture: author, timestamp, conversation, and content, exactly as the provider exposes it.

Per-conversation hash chains, so any later tampering with the archive is detectable.

Capture is read-only. Supervision records and reviews; it does not block or rewrite anything.

The review workflow

A queue your compliance team can actually work.

Flagged items arrive in a queue ordered by severity and rule. Opening flagged content requires a stated reason, and every access is logged and verified against the archive’s hash chain. Reviewers assign, escalate, and close each item with a disposition from your own policy, and every decision lands in an append-only audit trail with reviewer identity, timestamp, and the policy version in force.

Review item · audit trail

Representative example

09:12:04flaggedrisk rule: client communication
09:40:18assignedreviewer: compliance analyst
10:02:51content viewedreason recorded, access logged
10:09:33dispositionpolicy exception noted, attested

Custody

Your record, in storage you set the terms on.

Choose a Tallin-hosted archive on object-locked storage, written once and held immutably for the retention you set, or run the archive inside your own VPC, where content never leaves your boundary and only metadata and evidence egress. Review in the Tallin console is available for hosted deployments today; VPC-resident review is on the roadmap.

Not covered

NOT COVERED: Personal accounts and unsanctioned tools are not captured. Supervision covers the sanctioned enterprise workspaces you connect.
NOT COVERED: The archive holds what the provider's compliance API exposes, on the cadence Tallin polls it. Gaps are reported, never silenced.
NOT COVERED: Supervision records and reviews. It does not block, rewrite, or filter content in flight.

Supervision, specified

Sources: Claude Enterprise and ChatGPT Enterprise, through each provider's compliance API.
Archive: Message-level and append-only, with per-conversation hash chains. Hosted archives use object-locked storage; write once, immutable retention.
Access control: Flagged content opens only with a stated reason. Every access is logged, allowed or denied.
Review: A queue by severity and rule, dispositions from your own policy, and an attestation recorded with reviewer identity, timestamp, and policy version.
Retention: Configured by you, independent of the provider's own retention window.
Deployment: Tallin-hosted archive, or a customer-VPC archive where content stays inside your boundary and only metadata and evidence leave.

Put your sanctioned AI tools under supervision.

Connect a Claude Enterprise or ChatGPT Enterprise workspace, set your retention, and give your compliance team a queue they can work. Start with a $2,500, 60-day paid pilot, credited against year one if you continue.

Request access 2-minute self-serve scan — no signup