DATA & PROMPT HANDLING

What we store. What we don't.

Metering AI work means our systems can see prompts and responses. Here's exactly what is kept, where classification runs, how long content lives, and who can see it, in plain terms.

What we store

Metadata is the product.
Content is not.

📊
Metadata we keep
Model, token counts, cost, timing, identity, active tools, detected ticket references, and the derived work intent. This is what every report runs on.
💬
Content we hold briefly
A per-session transcript and the last captured message, used to attribute work to tickets and people. Held on the retention clock below, then removed.
🔒
What we never store
Your source code. Repository syncs record file paths and added/deleted line counts only, never file contents or diffs.
Classification & retention

In-boundary.
On a clock.

01
Classification stays local
Work-intent classification runs on a self-hosted model inside our boundary. Prompt excerpts are never sent to a third-party model.
02
Content on a retention clock
Captured prompt content is held for a configurable window (90 days by default), then removed. The reporting metadata is kept; the prompt text is not.
03
Hard delete, not a flag
When content ages out it is hard-deleted or redacted at the row level, not hidden behind a soft-delete flag.
Retention windowsConfigurable
Prompt content
90 days✓
Request metadata
365 days✓
Synced delivery data
730 days✓
Windows are set per deployment. When content ages out it is hard redacted. The session's intent, cost, and token metadata stay for reporting.
Isolation & access

Isolated by org.
Logged end to end.

01
Row-level tenant isolation
Each organisation's data is isolated at the row level and enforced in the database, not in application code alone.
02
Append-only audit trail
Every administrative and data action is written to an append-only, monthly-partitioned audit log.
03
People keep their identity
Each person authenticates with their own identity. Reporting leads with organisation- and team-level views; individual detail is opt-in, and we never produce individual performance rankings.
Access postureEnforced
Per-org isolation→ Row-level
Audit log→ Append-only
Classification→ In-boundary
Source code body→ Never stored
Get started today

Trust the numbers, because you can trace them.

Connect your tools and read your first AI spend readout today.

See your data freeRequest demo
14-day trial. No credit card. Setup in minutes, not weeks.