Holmes vs Momentic
Both use AI agents that explore your app, repair tests when the UI changes and check every pull request, and both price on usage, not seats. Momentic is built for developers who own the suite as YAML code. Holmes works from plain-language goals the whole team can read, and adds accessibility and API checks, context from designs and analytics, and a summary of every release.
Momentic is built for developers who write and maintain YAML specs in the repo. Holmes works from plain-language goals, so product and QA can write and review tests too, with no test code to maintain.
Holmes runs WCAG audits in a real browser, keyboard navigation included, and checks every request behind a journey. Momentic can call APIs from a test step, but dedicated API checks and WCAG audits aren't described on its pages.
Both read tickets and specs from Jira, Linear, Notion and Confluence. Holmes also reads the design in Figma and your top pages in PostHog before it decides what to test.
When something ships, Holmes writes what changed in plain language, then tests the journeys the change touched, so the whole team knows what went out.
Based on each product's public pages on 17 November 2026. A red cross means Momentic's pages don't describe it, not that it can't be done.
| Capability | Holmes | Momentic |
|---|---|---|
| Testing | ||
| Who it's built for | Yes: Engineering, QA and product in the same loop, with unlimited users | Note: Developers and DevX engineers who write and maintain the suite[7] |
| How tests are written | Yes: Plain-language goals. No test code to write | Yes: Plain-English prompts, a coding-agent skill or a web editor[3] |
| Where tests live | Yes: In Holmes, as plain-language goals. Existing test cases import from a file, Jira, Linear, Notion or Confluence | Yes: YAML specs in your repo; Mo opens a pull request with each new test[3][4] |
| Finds bugs nobody scripted | Yes: Agents explore and test paths nobody wrote down | Yes: Its Mo agent explores with parallel agents and reports each bug with reproduction steps[8] |
| Keeps tests working when the UI changes | Yes: Flows repair themselves when a label or layout changes | Yes: Locators re-resolve; lasting fixes arrive as an auto-heal pull request[9][10] |
| Web browsers | Yes: Chromium, Firefox and WebKit, the engine behind Safari | Yes: Chromium, Google Chrome and Chrome for Testing[11] |
| Native Android apps | Yes: Native Android apps on an Android emulator | Yes: Native Android, React Native, Expo and Flutter on local or hosted emulators[12] |
| Native iOS apps | Yes: Native iOS apps on iPhone and iPad simulators | Yes: On local or hosted simulators; out of beta[1][2] |
| Real devices | Not covered: Not covered. Android emulators and iOS simulators | Note: Simulators and emulators; its Mo page says human testers still cover physical hardware[1][13] |
| APIs | Yes: Every request behind a journey, checked in context | Partly: HTTP and GraphQL request steps inside tests; no standalone API suite[32] |
| Accessibility | Yes: WCAG audits in a real browser, with keyboard navigation | Not described: Not on its pages. WCAG audits |
| Workflow | ||
| Pull requests and previews | Yes: Check and comment on GitHub and Bitbucket pull requests. GitLab supported | Yes: A required check on pull requests; finds Vercel, Cloudflare Pages or Netlify previews[14][15] |
| Schedules and CI | Yes: On every deploy, on a schedule, or by webhook | Yes: On a schedule, or in GitHub Actions, CircleCI, Buildkite, Jenkins or another CI[14][16] |
| Where tests run | Yes: Holmes' cloud, in isolated containers destroyed after each session | Yes: Your own machines, your CI or Momentic-hosted infrastructure[5] |
| Context it reads before testing | Yes: Reads the code change, ticket, spec (Notion, Confluence), design (Figma) and analytics (PostHog) | Partly: Tickets and specs from Linear, Jira, Confluence and Notion. Figma or analytics aren't on its pages[17] |
| Coding agents (MCP) | Yes: Start runs and read findings from Claude and Cursor via MCP | Yes: CLIs and MCP servers for Claude Code, Cursor and Devin[18] |
| Evidence with each bug | Yes: Recording, screenshots, console and network logs, steps to reproduce | Yes: Repro steps, a screenshot or recording, video and trace, browser logs and network requests[19][20][21] |
| Flaky results | Yes: Multi-stage validation. Borderline issues re-tested by an independent evaluator | Yes: Flaky tests go to quarantine instead of blocking the pipeline[9] |
| Where findings go | Yes: Findings on the pull request, the ticket, Slack, Microsoft Teams and email | Yes: Bugs to Jira or Linear; Slack alerts; test cases sync to TestRail, Qase or Testiny[22][33] |
| Release summaries | Yes: Plain-language summary of what changed in each release | Not described: Not on its pages. A summary of what changed in each release |
| Pricing and security | ||
| What you pay for | Yes: Priced on how much Holmes tests, never on seats | Note: Usage: test steps, AI steps and browser or simulator minutes[23][24] |
| Users | Yes: Unlimited users on every plan | Note: Not priced on seats; invite the whole team[23] |
| Entry price | Yes: Free plan. Paid plans from €199 ($219) a month | Note: Free with $50 of usage a month; Scale $500 a month[25][26] |
| AI and your data | Yes: Zero-data-retention AI processing. Source code is never stored | Note: Zero-data-retention with its AI providers. Self-serve data can improve Momentic's own agents; Enterprise can opt out[27][28][29] |
| Inside your network | Not covered: Not covered. Cloud only | Partly: Control plane in Momentic's cloud; CLI execution can run inside your network[6] |
| Help getting started | Yes: Self-serve. Setup in under ten minutes | Note: Self-serve; on Enterprise its team helps you evaluate, migrate and roll out[30] |
| Company base | Yes: EU company (Ghent, Belgium), GDPR compliant | Note: San Francisco, California; Y Combinator W24[31] |
Holmes sells QA software, so we compete with Momentic. We've tried to be fair: every statement about Momentic comes from its own public pages, linked below with the date we checked them. Products change. If something here is out of date, email support@useholmes.com and we'll fix it.
Local or hosted iOS simulators and Android emulators. Hosted iOS can mock a camera feed.Checked 17 November 2026.
iOS and Android testing are out of beta.Checked 17 November 2026.
Authoring: plain-English prompts, the momentic-spec skill, or the web editor. Tests are YAML specs checked into your repo.Checked 17 November 2026.
Mo turns a flow you approve into a Momentic YAML test and opens a pull request in your repo.Checked 17 November 2026.
Run tests on your own machines, in CI, or on Momentic-hosted infrastructure.Checked 17 November 2026.
Not today. The control plane runs in Momentic's cloud. CLI execution can run inside your networkChecked 17 November 2026.
developers and DevX engineers who write and maintain the suite, not a separate QA org.Checked 17 November 2026.
It explores the app with parallel agents and returns a report of what it checked, what it couldn’t verify, and each bug with reproduction steps.Checked 17 November 2026.
Locators re-resolve when the UI moves, failure recovery patches the step, and permanent healing writes the fix back to the spec. Flaky tests go to quarantine instead of blocking the pipeline.Checked 17 November 2026.
An auto-heal pull request records the root cause, fix, and validationChecked 17 November 2026.
Web tests run on Chromium, Google Chrome, and Chrome for Testing; iOS on hosted simulators, Android on hosted emulators.Checked 17 November 2026.
Native Android, React Native, Expo, and Flutter apps on local or remote emulators.Checked 17 November 2026.
human testers still cover physical hardware and local product requirementsChecked 17 November 2026.
Run saved tests on a schedule, on every PR, or in CI, and use the result as a required check.Checked 17 November 2026.
Otherwise Mo checks the PR’s head commit once for Vercel, Cloudflare Pages, or Netlify preview links.Checked 17 November 2026.
Run repository tests in GitHub Actions, CircleCI, Buildkite, Jenkins, or your existing CI provider.Checked 17 November 2026.
Integrations such as Linear, Jira, Confluence, and Notion, plus custom MCP servers your workspace adds, give it tickets and specs.Checked 17 November 2026.
The momentic and momentic-mobile CLIs and MCP servers give Claude Code, Cursor, Devin, and other agents the ability to write and run real UI tests.Checked 17 November 2026.
Mo files each bug with repro steps, expected and actual behavior, and a screenshot or recording.Checked 17 November 2026.
The comment links each run back to the dashboard’s video, trace, and step evidence.Checked 17 November 2026.
Browser logs and network requests help you trace a UI failure to its cause.Checked 17 November 2026.
its test cases sync to TestRail, Qase, or Testiny, and an accepted bug files to Jira or Linear.Checked 17 November 2026.
No. Pricing is based on usage, not seats. Invite your whole team; you pay for what your tests use.Checked 17 November 2026.
$0.0125 / stepChecked 17 November 2026.
$50 of free usage each month.Checked 17 November 2026.
Scale costs $500 per month and includes $500 of usage, billed in advance.Checked 17 November 2026.
Momentic has zero-data-retention agreements with its AI providers.Checked 17 November 2026.
Data from self-serve accounts can be used to improve Momentic's own agents.Checked 17 November 2026.
Enterprise customers can opt out of using their data to improve Momentic's agents.Checked 17 November 2026.
Our team helps you evaluate, migrate, and roll out.Checked 17 November 2026.
Momentic was founded in 2023 and is headquartered in San Francisco, California. The company is a Y Combinator W24 company.Checked 17 November 2026.
Runs server-side without going through the browser. Combine with `saveAs` to feed the response into later steps.Checked 17 November 2026.
The Slack integration posts to your workspace when Momentic has something to reportChecked 17 November 2026.
The same job, step by step: from deciding what to test to telling you what broke.
Momentic
Mo reads tickets and specs from Linear, Jira, Confluence and Notion, plus any custom MCP servers you add.[17]
Holmes
Holmes reads the code change, the ticket, the spec in Notion or Confluence, the design in Figma and your top pages in PostHog.
Momentic
Each bug comes with repro steps, expected and actual behaviour and a screenshot or recording; accepted bugs file to Jira or Linear, and Slack gets an alert.[19][22]
Holmes
Each finding passes multi-stage validation, then arrives with a recording, screenshots, console and network logs and repro steps, on the pull request, the ticket, Slack, Teams or email.
They are close. Both use AI agents that explore your app, repair tests when the UI changes and check pull requests, and both price on usage. Momentic is built for developers: its tests are YAML specs in your repo. Holmes works from plain-language goals the whole team can read, reads designs and analytics as well as tickets and specs, adds accessibility and API checks, and writes a summary of every release.[3]
Momentic says it is built for teams that own quality in CI: developers and DevX engineers who write and maintain the suite, not a separate QA org. Holmes is built for engineering, QA and product together: tests are plain-language goals, and every plan has unlimited users.[7]
Holmes tests native Android and iOS apps on emulators and simulators. Momentic tests native iOS and Android apps, including React Native, Expo and Flutter, on local or hosted simulators and emulators. Neither runs on real devices; Momentic's Mo page says human testers still cover physical hardware.[1][12][13]
Both are priced on usage, not seats. Momentic's Free plan includes $50 of usage a month, Scale is $500 a month with $500 of usage included, and Enterprise is custom. Holmes has a Free plan with €250 in credits to start, Starter at €199 a month for up to 3 products, and Team from €499 a month, all with unlimited users.[25][26][23]
Momentic keeps tests as YAML specs in your repo and runs them on your own machines, in CI or on its hosted infrastructure. Holmes keeps plain-language goals in Holmes and runs them in isolated cloud containers, so there is no test code to maintain. You can import existing test cases into Holmes from a file, from Jira, Linear, Notion or Confluence, or from the top journeys in your analytics.[3][5]
Momentic re-resolves locators, patches failing steps and sends flaky tests to quarantine instead of blocking the pipeline. Holmes runs every finding through multi-stage validation, and an independent evaluator re-tests borderline issues, so what reaches your team is a bug worth reading.[9]
Momentic has zero-data-retention agreements with its AI providers; data from self-serve accounts can be used to improve Momentic's own agents, and Enterprise customers can opt out. Holmes uses zero-data-retention for all AI processing, never stores your source code, and runs in isolated containers that are destroyed after each session. See the Holmes Trust Center for details.[27][28][29]
If you want tests as code reviewed in your own pull requests, need to run tests inside your network, or want test cases synced to TestRail, Qase or Testiny, Momentic covers those on its own pages. If you want plain-language goals the whole team can work with, context from specs, designs and analytics, and accessibility and API checks in one run, try Holmes.[1][3][6][22]
Evaluating Momentic? See what Holmes finds first.
Free, no scripts, no SDK, and no test account needed to start.