For agencies · client-facing AI support agents
A sealed evidence report for every client agent you ship.
Keep each client in its own workspace, re-run their suite on a schedule or before a release, and hand them a dated report they can check for themselves — failures and unknowns included.
No card. About ten minutes to a first report.
- 1 workspace
- per client: agents, keys, retention, reports
- 5 roles
- owner, admin, operator, reviewer, auditor
- Daily or weekly
- schedules at a fixed UTC hour
- SHA-256
- lets a client check the report they hold
One workspace per client
Agents, policies, model keys, retention and reports stay apart — and every client's state is visible on one page.
Separate in the database
A row can only refer to rows in its own workspace — agents, policies, runs, suites — and that is checked in the database, not only in the interface.
Each report names its client
Rename a workspace and new reports carry the new name; sealed reports keep the name they were sealed with.
A portfolio view
Every client side by side with its latest result, how many scenarios failed or produced no result, and what is still running.
Invite the client in
As a reviewer to approve the scenarios you drafted for them, or as an auditor to read. Drafting and approving are different roles, and access changes are kept in an audit trail.
Keep testing after handover
A prompt changed on Tuesday and the client noticed first. Re-run before they do.
Schedules. A pinned suite version, daily or weekly at a fixed UTC hour, so runs compare like with like. A schedule that cannot start pauses and says why.
Release gate. A REST call starts a run; five exit codes tell a pass from a failure from missing evidence, and missing evidence stops the release too.
Signed webhooks. Each finished run is announced with counts and links — never a reply, an input, a policy or a key.
Rerun and compare. Fixed, newly broken, still failing, lost its result — and whether the agent's reply changed or only the graders did.
Production failures. Send in a real failure; it is stored redacted and comes back as a draft scenario a person approves.
Your key or ours. A client workspace with its own model key is graded on that key alone, and the report says who funded the grading.
When the client asks questions
The report you send is the one that was sealed. Your view of it travels beside it, never inside it.
Ready to share, or not yet
Before you send it, the run page says whether the report holds up — and why not, check by check. A report with failures can be ready: it is the evidence your client is owed.
Findings beside verdicts
Disagree with a verdict and record your own finding, with a reason. It sits next to the automated verdict and changes no count, grade or score.
Reissue with your review
A new sealed report carries the original unchanged, lists every verdict your team disagreed with and why, and names the original by its hash.
Withdraw a link at once
A report link expires after thirty days and can be withdrawn sooner. The link and every download stop working for everyone, and the withdrawal records who made it.
A reissued report says plainly that its reviewers are members of the workspace whose agent was tested, so a client reads your findings as yours.
What Novera is not
A document that overstates itself is worth less — especially to the person you hand it to.
Testing evidence, not certification
This report records Novera's testing of the declared agent, scenarios and configuration at the time shown. It is not a legal opinion, a conformity assessment, a certification or an assurance that any law is met, and Novera is not a notified body.
Not a certification. What you hand a client is evidence of testing, and the report says so on its cover.
Not legal advice. It does not say which obligations apply to your client.
One dated run. One suite version against one policy version, on the date shown.
Only agents you may test. You confirm you own each agent or are authorised to test it; that confirmation is sealed into its reports.
Every report carries the same notice and its own scope and limitations. Read the limits in full.
Run it against a client's agent
Three free runs for your account, shared across every client workspace you own. Several clients already? Ask for a pilot and a person replies.
- No card
- Email, Google or GitHub
- Database in Frankfurt (EU)