Momentic tests your app the way a user would. You describe a flow in plain English, such as "click Checkout, type the email, check that an order confirmation shows", and an AI agent carries it out in a real browser, iOS simulator or Android emulator. When the UI changes, the test adapts instead of breaking.
The pitch is simple: AI coding tools write code faster than teams can check it, so the checking needs AI too. The site's tagline says it: "Catch bugs before they ship."
| Company | Momentic, San Francisco, Y Combinator (W24) |
| Founders | Wei-Wei Wu (CEO, ex-Density) and Jeff An (led testing at Robinhood and enterprise quality at Retool) |
| Funding | About $19M: YC pre-seed, a $3.7M seed in 2025, and a $15M Series A on 24th November 2025 led by Standard Capital, with Dropbox Ventures |
| Platforms | Web on Chromium, iOS simulators, Android emulators. No physical devices |
| Customers | Notion, Quora, Webflow, Retool, Runway, Xero and others |
| Security | SOC 2 Type 2, SSO and role-based access |
What it does
- Natural-language tests. Each step is plain English. No CSS selectors or XPath to maintain.
- Self-healing. When a button moves or gets renamed, the AI finds it again and updates the step. It's meant to tell an intended UI change from a real regression.
- Tests as code. Tests are YAML files in your own repository, reviewed and versioned like any other code.
- Mo, the AI QA engineer. A separate agent that explores the app on its own, reads your docs, Jira, Linear or Figma, maps the user flows and reports gaps in coverage.
- Failure analysis. A failed run comes with steps to reproduce, a session replay and a root cause. Auto-maintenance can open a pull request with the fix.
- CI gate. 1 CLI command runs the suite on a laptop, in CI or in a coding agent's sandbox, and blocks a pull request or deploy that breaks a flow.
For business people
The problem: end-to-end tests are the tests that click through the app like a customer. They catch the bugs that hurt most, but they're slow to write and break every time the UI changes. Many teams give up on them and test by hand before each release.
What Momentic changes: writing a test takes a sentence per step, and keeping it alive takes much less work. The customer stories on the site show the effect, in the vendor's own words:
- Quora: daily test runs down from 7 hours to 30 minutes.
- Retool: 8 times more releases.
- CoverGo: 30% fewer production incidents.
Pricing
Momentic charges for usage, not per seat. 1 credit is 1 test step, and an AI-driven step costs 2.
| Plan | Price | Includes |
|---|---|---|
| Free | $0 | 2,000 credits a month (about 200 test runs), 5 parallel browsers, 1 mobile device |
| Pay-as-you-go | $125 a month | 10,000 credits (about 1,000 runs), then $0.01875 a credit, 25 browsers, 5 iOS and 5 Android devices |
| Enterprise | Custom | SAML SSO, SCIM, audit log, SLA, longer mobile sessions |
The limits:
- AI in the loop means cost and some non-determinism. Every AI step spends credits, and an agent can read a screen differently from 1 run to the next.
- No real devices. Mobile tests run on simulators and emulators only.
- A young vendor. Founded in 2023, with a Series A behind it. Tests stay as YAML in your repository, which limits the lock-in.
For technical people
A test is a YAML file with an id, a start url and a list of steps. This example is from the docs:
fileType: momentic/test/v2
id: silver-river-lantern
url: https://shop.example.com
steps:
- click: Checkout
- type:
text: jeff@example.com
into: Email input
- assert: An order confirmation is visible
Setup is a wizard that installs the CLI, writes momentic.config.yaml and a sample test, installs Chromium and runs the sample:
npx @momentic/wizard@latest # interactive setup
npx momentic run add-to-cart --upload-results # run 1 test
npx momentic lint # validate test files
npx momentic install-mcp # MCP server for your coding agent
- Runtime: Node 22.12 or 24. Web tests run on Chromium.
- Coding agents: a local MCP server gives Claude Code, Cursor, Codex and others a live browser or device plus tools to write, run and triage tests.
npx momentic skillsinstalls agent skills into the repository. - Mo ships as its own CLI,
qa(npm install -g qa), withqa start "<brief>"to begin an exploratory session. - Extras: JavaScript steps in a sandboxed Node environment or in the page, dedicated phone numbers for SMS and OTP flows, and an analytics REST API.
- Docs for agents: the site publishes llms.txt, which tells an agent not to invent step names, config keys or CLI flags.
How it compares
| Tool | Approach |
|---|---|
| Playwright | Free, open source, fast, but tests are code with selectors that you maintain |
| Selenium | The older standard, same selector model (see Python Library: Selenium ) |
| Momentic | Plain-English steps, self-healing, paid per step |
Value
My lab apps are built mostly by coding agents now, and nobody clicks through them before I use them. A test that says "log in, open the dashboard, check the pipeline table loads" in plain English is something an agent can write, run and fix with me out of the loop. The MCP server and the skills make that the default workflow, not an add-on.
The free tier covers about 200 runs a month, which is enough to try it on 1 app. For a solo builder, Playwright driven by Claude Code does much of the same for free, so the test is whether self-healing saves more time than the credits cost.
Further Reading
