Test management + AI evals

Every test, run and defect in one calm place.

No more spreadsheets and scattered tickets. Plan, run and track testing together.

Then ask the harder question.

Is your AI assistant safe to release? Install a ready-made eval pack and find out.

Ship when it is ready. Not before.

EvalNest is test management with ready-made eval packs built in.

Test management + AI evals

Ship when it is ready. Not before.

Test cases, runs, defects and ready-made AI eval packs, all in one calm place.

01 / Test management

Everything your QA team tracks, finally together.

Test cases, runs, defects and releases live in one place, so nobody has to ask which spreadsheet is current.

  • Repository: suites and test cases with steps, priority, type and automation status.
  • Test runs: assign testers, record passed, failed or blocked, tied to staging or production.
  • Defects: log one from a failed result, then push it to Jira or GitHub in one click.
  • Releases and coverage: see which stories have tests and whether a release is ready.
  • Test review: changes to test cases get approved before they go live.
  • Automation: results arrive from CI (JUnit, Playwright, Cypress) and a CI step can fail the build.
Projects / Repository
EvalNest repository with suites and test cases showing priority, type and automation status
AI Evals / Overview
EvalNest Evals overview with ready-made packs and pass rate by category
Release gate
Release gate blocking a release, with the reasons listed
02 / AI evals

Ready-made tests for the AI assistant you are about to ship.

Install a pack, connect your assistant, and see whether it is safe to release.

  • Prompt injection
  • Data privacy
  • Invented facts
  • Unsafe actions
  • Hand-off to a human

Packs for 11 industries plus general packs, including banking, healthcare, insurance and retail.

Results map to OWASP LLM Top 10, NIST AI RMF and the EU AI Act.

CI can run the release gate on every build, so a risky change never ships quietly.

03 / Release gate

Let the release land only when it is safe.

Each run passes or blocks the release against your rules, and the approval is recorded. Hold the button to run a sample gate.

  • No critical failuresWaiting
  • Pass rate at or above 90%Waiting
  • Prompt injection at 100%Waiting

A sample gate with demo results. Your rules, your thresholds.

04 / Product Eval Pack

Validate your product against its own stories and requirements.

The Product Eval Pack turns what your product is supposed to do into checks that actually run. Bring a Jira story or a requirement, and its acceptance criteria become tests, with a defects list at the end.

  1. 1Your story or requirement
  2. 2Its acceptance criteria
  3. 3Browser tests that run
  4. 4A defects list with evidence
One of the packs inside

The Web Eval Pack: 136 checks for any web app, run in a real browser.

136

checks across 12 kinds of testing

101

run automatically

7 of 7

planted bugs found on our sample web app, in about 3 minutes

The 12 packs in the Web Eval Pack
PackWhat it checksChecks (automatic)
SmokeApp loads, signs in, no raw or script errors, quick accessibility and speed check10 (10)
Sign-in & SessionsWrong passwords, empty fields, sign-out, private pages, session cookies18 (10)
Access RightsA lower-rights account cannot open admin pages or other users' records5 (2)
Navigation & LinksBroken links, back and forward, refresh, direct addresses10 (8)
UI & VisualBroken or stretched images, clipped text, leftover placeholder text10 (6)
ResponsivePhone, tablet, landscape and wide screens8 (7)
Cross-BrowserChrome, Firefox, Safari (WebKit), Edge, phone browsers7 (5)
AccessibilityWCAG 2.2 AA: scan, keyboard, focus, contrast, zoom16 (13)
PerformanceCore Web Vitals, page weight, slow 3G, backend speed14 (11)
Security BasicsHTTPS, security headers, exposed files, XSS, certificate10 (9)
Forms, Files & ErrorsRequired fields, formats, limits, long text, double-submit14 (9)
SEO, Privacy & ReleaseSearch tags, trackers before consent, privacy links, cache-safe releases14 (11)

Rule-based, with no AI model and no cost per run. Safe mode never clicks Delete, Pay or Send.

Eval packs / Web Eval Pack
The Web Eval Pack opened, with 12 packs to install and run
The Web Eval Pack: 12 packs to install and run.
Eval runs / Release 2.4
Results by pack for a full run on the sample web app
Results by pack for a full run on the sample web app.
How a Web Eval Pack run works

Five steps from install to a list of real defects.

  1. 1

    Install

    Install the packs you need. Each adds its checks as normal test cases.
  2. 2

    Pick

    Click Run Web packs, then pick the web app and the packs.
  3. 3

    Explore

    EvalNest signs in with a test account, explores the pages and runs every automatic check.
  4. 4

    Review

    Results show passed, failed or blocked per pack. Every defect has a screenshot and steps to reproduce.
  5. 5

    Hand over

    Checks that need a person, like screen-reader testing, are listed for your team.
05 / Works with your AI assistant

Ask in plain words. EvalNest does the work.

EvalNest includes an MCP server, so Claude, Cursor and other assistants connect with a personal API token.

  • Write test cases for this story.
  • Start a regression run on staging.
  • Log a defect for the failed checkout test.
  • What does our coverage look like for release 4.2?
  • Run the banking eval pack on Assistant v2.

The assistant has your permissions, nothing more, and everything it does goes in the audit log.

Profile / AI & MCP server
Connecting an AI assistant through the EvalNest MCP server
A closer look

Nine screens, one calm place.

Repository: suites and test cases with priority, type and automation.
A test run with results, progress and pass rate.
Quality dashboard across projects.
AI Evals overview with the ready-made packs.
Release gate blocking a release, with the reasons.
The Web Eval Pack: 12 packs to install and run.
Results by pack for a full run on the sample web app.
Defects with screenshots and one-click logging.
Connecting an AI assistant through the MCP server.

Screenshots show fictional demo data.

Questions

The things teams ask first.

No. Defects push to Jira or GitHub in one click, and releases and stories stay linked.

No test can prove that. A passing run is strong evidence against the risks it covers, mapped to OWASP, NIST and the EU AI Act. Rerun it on every build so changes cannot slip past.

No. It is rule-based, with no AI model behind it and no cost per run.

Safe mode never clicks Delete, Pay or Send.

Yes. Results come in from JUnit, Playwright and Cypress, and a CI step can fail the build or run the release gate.

Exactly what the person who connected it can do. Every action is recorded in the audit log.

Book a demo

Give your next release a safe place to land.

Book a 30 minute walkthrough with your own test process in mind.

Request received.