Control high-accountability AI workflows without losing the operating record.
Turnkeeper brings policy records, human review, and append-only lifecycle evidence into one organization-scoped workspace for AI-agent operations.
Design-partner access. Production Replay remains gated.
- 01Agent eventRequest receivedMetadata accepted
- 02Policy recordContext preservedReference attached
- 03Human reviewReview openedState recorded
- 04Decision evidenceOutcome retainedEvent appended
- organization
- tk_enterprise
- project
- support_ops
- review_state
- open
- privacy
- metadata_only
Replace fragmented oversight with one durable record.
Enterprise AI operations break down when policy context, review decisions, and event history live in different systems.
- 01Policies live away from the work
Keep workflow policy records and defined decision states in one operating surface.
- 02Reviews happen in side channels
Track the proposed action, policy reference, priority, and review state together.
- 03Evidence is reconstructed later
Capture ordered, append-only lifecycle metadata by project as work happens.
- 04Ownership becomes ambiguous
Partition records by organization, project, and environment; bind credentials to projects.
A shared operating layer for consequential workflows.
Four connected surfaces keep responsibility and evidence legible.
- 01
Policy records
Keep the governing reference close to the workflow and the resulting decision.
- 02
Human review
Give operators a defined queue, priority, and durable review state.
- 03
Replay evidence
Retain ordered lifecycle metadata without storing message content in Replay.
- 04
Scoped access
Separate organizations, projects, environments, and project-bound credentials.
Move reviews out of side channels.
Keep the proposed action, policy reference, priority, and decision state in one review context so operators can act with the same record.
- Explicit review states
- Defined decision states
- Linked policy references
| Request | Policy reference | Priority | State |
|---|---|---|---|
| Refund exception | refund-exception-v3 | High | Open |
| Account closure | account-change-v2 | Medium | Approved |
| Sensitive disclosure | disclosure-v1 | High | Blocked |
- Selected request
- Refund exception
- Proposed action
- Requires review
- Decision state
- Open
Keep lifecycle evidence ordered and inspectable.
Replay accepts strict, metadata-only lifecycle events and retains them as an append-only project history for investigation and operational review.
- Canonical lifecycle event types
- Idempotent ingestion
- Project-scoped event history
- 001input.received
- 002turn.started
- 003turn.decision_recorded
- 004output.held
- event_id
- evt_01J7C2P
- project_id
- support_ops
- privacy.mode
- metadata_only
Start with one governed workflow.
A focused design-partner pilot can move from scope to evaluation in 30-45 days.
- 01
Scope
Choose one consequential workflow, its owners, and the decision boundary.
- 02
Map
Define the policy references, review states, and evidence the team needs.
- 03
Connect
Instrument lifecycle metadata with project-bound credentials.
- 04
Evaluate
Review the operating record, adoption path, and production readiness together.
Clear boundaries before broader deployment.
We describe the product by its current operating boundaries, not by a future-state checklist.
Technical resources
Inspect the product surface and current public availability before a pilot.
See the shortest path from a project credential to a recorded lifecycle event.
02Replay APIReview the metadata-only ingestion contract, event types, and response behavior.
03Public source statusCheck exactly which developer surfaces are public, pre-release, or gated today.
04SupportSend product, implementation, account, or security questions to the team.
Build the operating record around one real workflow.
Bring the workflow, its decision boundary, and the evidence your team needs. We will scope the pilot together.
Request an enterprise pilot