Everything Verirun does

Built for developers on Vercel, Supabase or Railway whose coding agent says it is done. Verirun maps your app, runs the flow in a real browser, and returns PASS or FAIL with what the browser saw and what your backend logged. The flow stays as a test that reruns for free. It works on any URL; backend logs need a connected log source.

verirun record <name>

Record a flow by doing it once

A browser opens on your app. You do the flow by hand and close the window. Each click and keystroke is saved with the selector Playwright used, so replay targets exactly what you clicked.

  • --url where to start
  • --channel chrome record in your installed Chrome or Edge
  • --user-data-dir a signed-in browser profile, for flows behind a login
.verirun/workflows/invite-member.jsonExample file
{
  "schemaVersion": 1,
  "name": "invite-member",
  "browser": "chromium",
  "steps": [
    { "index": 0, "action": "navigate",
      "params": { "url": "http://localhost:3000/dashboard" } },
    { "index": 3, "action": "fill",
      "params": { "text": "grace@example.com" },
      "selector": "internal:role=textbox[name=\"Email address\"i]" },
    …
  ]
}
verirun run <name>

Replay it after every change

Headless, step for step, in the browser you recorded with. You get PASS or FAIL first. When a step fails you get the step, the error, the page and a screenshot, then the failing API call joined to the backend line that explains it.

  • --headed watch the replay in a visible window
  • --cdp attach to a Chrome you already have open, and leave it open
  • --timeout how long each step may wait
TerminalExample output
verirun run profile --timeout 5000
FAIL profile step 4 click "Back to dashboard" (3/4 replayed, 5.6s)
  locator.click: Timeout 5000ms exceeded.
  url: http://localhost:3000/settings
  screenshot: .verirun/runs/profile-2026-11-14T09-12-41-388Z/failure.png
step 3 click "Save" → PATCH /api/profile 500 → vercel: column "display_name" does not exist [i1]
verirun log vercel (or supabase, or railway)

See what your backend did during the run

A browser test only sees the screen. Connect the log source your backend runs on, once, and every run pulls the logs from its own time window and prints the errors under the result. Pick one; you do not need all three.

VercelRuntime logs of the deployment under test.
verirun log vercel
SupabasePostgres, edge function and auth logs.
verirun log supabase
RailwayDeployment logs of one service.
verirun log railway

--user ada@example.com keeps only the lines about one user. The full set goes to logs.json next to the run.

What the browser saw
Invoice INV-1043 sent
Globex · $4,800 · Sent
What the backend logged in those 1.8 seconds (example)
PASS send-invoice 6/6 steps 1.8s

x2 vercel/lambda error
    POST /api/invoices/send
    email provider returned 422

vercel/lambda error
    duplicate key value violates
    "invoice_events_pkey"
verirun agent --skill --install all

Let Claude Code, Codex or Cursor run it

One command installs a skill your agent loads when a task looks like testing. From then on it runs your saved flows itself, branches on the exit code, reads the failed step and the backend errors, and checks its own fixes in a real browser.

  • --install claude, codex, cursor or all
  • verirun workflow check validates a flow the agent wrote, without a browser
  • Any other agent can read the output of verirun agent
What the agent runsExample output
verirun run profile --json
{"status":"failed","steps":{"total":4,"replayed":3,"failedAt":3},
 "logs":{"problems":1, …}}
exit 1

# the agent fixes the code, then checks its own fix
verirun run profile --json
{"status":"ok","steps":{"total":4,"replayed":4}, …}
exit 0
verirun map

See every page and API call, from your source

Static analysis of your React or Next.js code, with no browser and no running server. It finds the routes, follows components across files, and links each button and link to the page it opens or the API it calls.

  • --json a Playwright selector and suggested action for every element, so an agent can write flows
  • --bfs /settings what is reachable from one page
  • --reach how much of each page it captured, so gaps are visible
verirun map, drawnExample graph
/dashboardsidebar/projectssidebar/invoicessidebar/teamsidebar/settingsCreate projectPOST /api/projectsSend invoicePOST /api/invoices/sendSend invitePOST /api/invitesSave changesPATCH /api/profile
page backend call navigation effect, labelled with the element
verirun scan <url>

Check a live site from the outside

Up to 20 pages in headless Chromium, several at a time, two links deep. It reports JavaScript exceptions, console errors, failed requests, dead links and anchors that point nowhere. It needs no source code.

It only reads: GET and HEAD, never a click or a form, robots.txt obeyed. The scanner on the home page is a quick version of it you can try now.

TerminalExample output
verirun scan https://fieldbook.example
FAIL  https://fieldbook.example/  3 findings on 12 pages (browser, 4.2s)
JS exception TypeError: Cannot read properties of null /pricing [f1]
HTTP error 404 GET /static/og-image.png / [f2]
dead link link to /changelog returns 404 /docs [f3]
+1 hint (--verbose)
--json · exit 0 / 1 / 2

Made for scripts and CI

Every command exits 0 when the check passed, 1 when it failed and 2 when it could not run, and prints a versioned JSON result with --json. In CI, a license key in an environment variable replaces the browser login.

CI stepExample output
# VERIRUN_LICENSE_KEY is set as a CI secret
npm install -g @verirun/cli
verirun workflow check invite-member
verirun run invite-member --json > result.json
# exit 0 passed, 1 failed, 2 could not run

It runs on your machine. Your data stays there.

The browser, your recordings, the logs it pulls and your API tokens live on the machine you run it on. Log queries go straight from there to your provider.

The one thing the CLI sends to us is a license check. Details are in the privacy policy.

Your machine
  • Browser and recorded flows
  • Run logs (logs.json)
  • API tokens (.verirun/config.json)

Questions

Everything else is in verirun --help, man verirun and the README.

Do I have to write test code?

No. You click through the flow once and the steps are saved as JSON. You can edit that file, or have an agent write one from verirun map --json, but you never have to.

How is this different from Playwright tests?

Verirun runs on Playwright, so the browser side is the same engine. The differences: you record instead of writing code, and each run also reads your backend logs for its own time window, so a test that passes on screen still shows you the error behind it.

Is there AI in the replay?

No. Replay feeds the recorded selectors back to the browser in order. No model judges the page, so the same recording against the same app gives the same result, and a run has no AI cost.

What happens when my UI changes?

Replay uses the exact selector you recorded, so a renamed or removed element fails that step and the output names it. Re-record, or have your agent write a new workflow from the map and confirm it with a run.

Do I need Vercel, Supabase and Railway?

No. Connect the one your backend runs on, or none. Without a log source, runs still replay and pass or fail; they just have no backend lines to show.

Which stacks does it support?

Record, run and scan work on any web app you can open in a browser: localhost, a preview, staging or production. verirun map reads TypeScript React apps using react-router or the Next.js App Router.

Can I run it in CI, or offline?

In CI, set VERIRUN_LICENSE_KEY and branch on the exit code (0 passed, 1 failed, 2 could not run). Offline, the signed license keeps working; after 15 days without a check it starts reminding you, and after 30 it stops until it can check in.

Try it on your own app

3 days free, no card. Then pay once for lifetime access.