Finding your way around Novera

This page walks through Novera's screens once you have signed in: what each place is for, what you do there, and where to find the report you hand to a client. For what each part does and why, see Novera in plain words.

The order of the work

Novera's navigation follows the order the work happens in: connect an agent, run a test, share the sealed report. The top bar holds five places, in that order:

• Dashboard: where things stand, and the one thing to do next.
• Agents: the AI agents you test.
• Scenarios: what the agents are tested on.
• Runs: every test, what it found, and the reports.
• Settings: your profile, the workspace, its members, keys and billing.

On a phone or a narrow window, these are inside the menu button at the right of the top bar.

How you use Novera changes the labels

When you sign up you say how you will use Novera, and you can change it later under Settings → Profile and preferences. Nothing you have moves when you change it.

• Personal: for testing your own agent. Agents and Runs are called My agent (or My agents) and My runs.
• Agency: for testing agents you deliver to clients. Each client gets a workspace of their own, so their agents, keys and reports stay apart, and each report names that client. A switcher next to the logo, headed Clients, moves between them; the dashboard lists your clients.
• Enterprise: for several teams. The top bar adds Workspaces and, for those allowed to see it, Audit.

The top bar

From left to right:

• The Novera logo goes back to the home page.
• The workspace switcher, when you belong to more than one workspace or use agency or enterprise mode.
• The five places above.
• A short label saying who pays for grading in this workspace: Novera's trial (with runs left), a pilot (with runs left, shared across the workspaces its owner owns), a plan, or your own model key. The owner and admins can open billing from it.
• Your role, if you are not the owner.
• Run evaluation (shortened to Run on smaller screens): start a test from any page. Choose the agent, the suite and, where the suite offers it, the language.
• The menu: your name and role, Connect an agent, Members, Billing, Your profile, the Step-by-step guide, Novera in plain words, the Documentation, and Sign out. What appears depends on your role.

At the bottom right of every signed-in page is Ask Novera, an assistant that answers questions about Novera from this documentation and about your own runs from their stored counts. It cannot change anything. See The Ask Novera assistant.

Dashboard

Before your first test, the dashboard shows the getting-started steps: connect your agent, write its policy, run the suite. Each step ticks itself only when it has actually happened. If you have no agent address to hand, Or try a sample support agent first adds Novera's own test agent.

Afterwards, it shows one sentence about the state of the workspace, three figures (the latest run, open findings, reports ready to share), Needs attention (at most five things that need someone, each with a link to where it is resolved), and your recent runs. The button at the top right is always the most useful next step for your role.

Agents

The list of agents in the workspace. Connect an agent adds one: see Connecting your agent, step by step. Open an agent to find its tabs:

• Overview: its recent results.
• Policy: the rules the agent should keep. Saving creates a new version; old versions are kept.
• Runs: every test of this agent.
• Schedule: run a suite automatically every day or week. See Scheduled runs.
• Connection: the receipt of the last connection check, Change the connection, Production or test target, Verifying what the agent does (the optional read-back), Gate pull requests in CI, and archiving.

The Run the suite button at the top of an agent's page starts a test. If the agent has no policy yet, the button is Write the policy instead. Under What are you testing? (optional) you can note the agent release and knowledge-base version; the report shows them as declared by you.

Scenarios

The tabs at the top:

• Library: the suites your workspace can run, and their scenarios.
• Catalog: every built-in suite and every version of it, with what changed between versions. From here you can also try one scenario alone against one of your agents.
• Drafts: scenarios waiting for a person to approve or reject them, and approved ones waiting to be added to a suite. This is also where you can run up to five approved scenarios beside an official suite (Extend an official suite).
• Draft or import: draft scenarios from your policy, or import them from a file or another testing tool. See Writing and importing a suite.
• Builder: the Suite Builder. See The Suite Builder.
• Regressions: turn a mistake your agent made with a real customer into a scenario. See Regressions from production.

Runs

The tabs at the top:

• All runs: every test, filtered by Failed, Evidence incomplete, Passed or In progress.
• Review: what needs a person across the last ten finished runs, grouped as Needs attention, Evidence incomplete, New regressions, Unverified actions, Awaiting approval and Closed.
• Reports: every sealed report, with whether it is ready to share.
• Compare: choose two runs and see them side by side.

A run's own page

While a run is in progress, its page shows each scenario as it is recorded. The result appears only when every scenario is recorded, never before. Keep the page open: in the app, the open page is what moves the run along.

If a run has not moved for a few minutes, the page says so: "This run has not moved for 4 minutes. Nothing is lost: every scenario recorded so far is kept." Press Continue run to carry on. If another process is already advancing it, the page says that instead and the run continues on its own. A run that records nothing for 24 hours is stopped, and no report is sealed for it.

When the run is finished, the page shows:

• The scorecard: the grade, the counts and how the verdicts were reached.
• The client report bar, when a report was sealed: its readiness (Ready to share, Review before sharing, Incomplete, Grade withheld, and so on), View client report, a button to copy the link, Print / save PDF, Verify seal, Compare with previous run when there is one, and Downloads (Markdown, CSV, JSON, JUnit). A run against Novera's test agent is labelled test data here too.
• How to read this run (60-second guide): a closed box with the same plain explanation the client report carries.
• Rerun and compare against this: start the same test again, compared with this one.
• Five tabs: Overview (what to look at first), Cases (every scenario, filtered by Failed, No result or Passed; open one for its full evidence), Comparison (against an earlier run of the same suite), Evidence (the seal, the inputs recorded before the run, Before you send it and Withdraw this report) and Activity (what happened to this run, and when).

On a failed scenario, Why did this fail? asks for a proposed change to your policy, and Try this scenario alone sends that one scenario again. On any scenario you can record your own finding beside the verdict. See How a run works.

Sharing, printing and checking a report

• To share: on the run's page, copy the link from the client report bar. The link is private, expires after 30 days, and can be withdrawn from the Evidence tab.
• To print or make a PDF: Print / save PDF opens the client report and your browser's print window; choose "Save as PDF" there. The printed report keeps its seal and a short reading legend on the first page.
• To check the seal: Verify seal opens the public seal check. Anyone else can use the seal check too, by pasting the fingerprint or dropping the JSON download.
• To compare: Compare with previous run, or Runs → Compare.

For how a client reads it, see Reading a report.

Settings

On the left (at the top on a phone), grouped by whom they concern:

• You: Profile and preferences (how you use Novera, your preferences, what Ask Novera remembers, and downloading your own data or deleting your account) and Sign-in and security (password, sign-in methods, signing out everywhere).
• This workspace: General (its name, exporting the whole workspace, how long the agent's raw replies are kept, and erasing or leaving the workspace), Grading and data routing (which models grade your runs, connecting your own model key, and what kind of data each grading provider may receive), Members (invite people and set roles; shown once there is more than one member, or in agency and enterprise mode) and Billing.
• Developer: API keys, MCP and webhooks.
• Security: Audit log, for the owner, admins and auditors.

The ? explanations

A small ? next to a word or a figure opens a short explanation in place: what it is, why it matters, and what to do next, often with a link to the page of the documentation that says more. It opens with a click or the keyboard and needs nothing else. The ? marks are left out when a page is printed. For every word in one place, see the plain-words glossary.