Time Machine
Time Machine lets you replay any past AI session in full context: what the model received as input, what it returned, which governance annotations were attached, what risk level was assigned, and what the outcome was — all as a point-in-time snapshot.
This is useful for three things:
- Incident investigation — understand exactly what happened in a flagged session
- Audit and compliance review — present a complete, unambiguous record of a specific decision
- Model evaluation — compare behavior of a past model version against a new one on the same inputs
Opening Time Machine for a session
- Navigate to Decisions in the Customer Portal.
- Search for the session by ID, date, outcome, or any other filter.
- Click the session row to open the detail view.
- Click Time Machine in the session detail toolbar.
Time Machine opens in a new panel with the session record loaded at its original state.
What you see in Time Machine
The Time Machine view has three panes:
Session timeline
A chronological event trace of the session, showing each instrumented step with its start time, relative duration, and governance tags. You can click any step to expand its recorded context.
Decision snapshot
The complete DecisionContext as recorded at ingest time:
| Field | Value at time of decision |
|---|---|
| Input summary | Recorded input trace (if content capture was enabled) |
| Intent | Model-classified intent label |
| Model | Model identifier and provider |
| Decision | Approved / Denied / Deferred |
| Confidence | Confidence score at decision time |
| Risk level | MINIMAL / LOW / MEDIUM / HIGH / CRITICAL |
| Outcome | Actual recorded outcome (set after the fact) |
Governance context
All governance annotations recorded on the session, including:
- Guardrail actions (blocked, allowed, flagged)
- Human oversight indicators
- Grounding status
- Content safety flags
- Agent role
- Governance score
Reconstructing context for a model update
When you upgrade a model and want to understand behavioral differences, Time Machine lets you load a past session and see the original context alongside the new model’s response.
Time Machine shows recorded state — it does not re-run the model against the original inputs. It reconstructs what was captured by the SDK instrumentation at the time. If content capture was disabled (the default), input and output text will not be present in the snapshot.
Enabling content capture for future sessions
Time Machine is richest when content capture is enabled. To enable it:
configure_veriproof(
api_key=os.environ["VERIPROOF_API_KEY"],
application_id="your-app",
capture_content=True, # Includes input/output in the session record
)Content capture should only be enabled after reviewing your data handling obligations. Input and output text may contain PII. VeriProof stores captured content encrypted at rest and access-isolated per tenant. See Multi-Tenant Isolation for storage details.
Sharing a Time Machine view
You can share a direct link to a Time Machine session view with other portal users in your account:
- Open the Time Machine view for the session.
- Click Copy Link in the toolbar.
- Paste the link and share it.
The link is only accessible to portal users who are logged in to your account. It is not publicly accessible.
Exporting from Time Machine
To export the full session record for compliance or legal review, click Export Evidence from the Time Machine toolbar. This opens the same evidence export workflow as described in Compliance Evidence Export.
FAQ
Does Time Machine show sessions from the sandbox environment?
Yes. When you are viewing sandbox sessions (via the sandbox toggle in the portal), Time Machine behaves identically. However, sandbox sessions are not blockchain-anchored, so the anchor section will show “Sandbox — no blockchain anchor.”
How far back can I go?
Session records are retained according to your plan’s retention period (2 years for Starter, 7 years for Growth, configurable for Enterprise). Time Machine can access any session within your retention window.
Time Machine shows some fields as “Not recorded.” Why?
This means the SDK did not capture that field for that session. Common causes:
- Content capture was disabled (inputs/outputs will not appear)
- The governance annotations were not added by your application code
- The session predates a schema addition (e.g.,
outcomeis only available if you calledsession.record_outcome()in your application)
Next steps
- Compliance Evidence Export — export the Time Machine view as a regulatory package
- First Integration — add governance annotations so Time Machine captures richer data
- Governance Scoring — understand the score shown in the governance context pane