Can people pay and get in?
ok POST /api/signup
fail POST /api/checkout
Recognised by name, or listed in prodbat.json.
Routes matched by name (checkout, billing, login, signup) or declared under "flows" in prodbat.json.
savetheworld.comis working.needs a look.is broken.
Everything is working.All checks passing · 3/3 pages 200 · p95 412 ms
Working, and one thing needs a look.1 advisory · next@14.2.3 · CVE-2025-29927 (CVSS 9.1)
People cannot pay.POST /api/checkout 5xx rate 64% (9/14 req, 15m)
next 14.2.3 has a security fix (14.2.25). Your agent can update it.
next@14.2.3 → 14.2.25 · middleware authorization bypass · package-lock.json
Monitoring for apps built with AI
Prodbat watches your live app day and night: whether it loads, whether people can sign in and pay, and whether anything is crashing or out of date. When something breaks, you get an email in plain English and your coding agent gets the steps to fix it. Then Prodbat checks the live app again to confirm the fix worked.
Synthetic checks every 5 min (1 min from 3 regions on Pro). Server 5xx and browser errors with file:line, sign-in and checkout failures raised as Broken, cron and queue heartbeats, and a daily lockfile scan against OSV.dev and endoflife.date. Alerts go out by email and Slack. Your agent pulls each one as a task with npx prodbat next, and npx prodbat done re-checks from outside before the task closes.
Open source. Read every line on GitHub, or run it yourself for free.
Install Prodbat on this project. Run npx prodbat init, do what it prints, deploy, then run npx prodbat verify.
The fix usually takes minutes. Winning back the customers who found the bug before you did is the hard part. Prodbat watches the live app for you, tells you what broke in a sentence, and gives the fix to the agent that wrote the code.
Each check shows its industry name and the alert Prodbat would send you about your app. Want the numbers? Switch to Engineer, bottom right.
"Your sign-in page is not loading. The home page loads."
GET /, /login, /pricing every 5 min (every 1 min from 3 regions on Pro; regions must agree). TLS expiry at 14 d, domain at 30 d.
"People cannot pay." One failed checkout or sign-in is enough to alert you.
Browser errors, failed fetches and server 5xx with file:line. A single failure on a pay or auth route opens a Broken incident.
Recognised by name, or listed in prodbat.json.
Routes matched by name (checkout, billing, login, signup) or declared under "flows" in prodbat.json.
"The job that saves the world every night hasn't run since Tuesday."
Heartbeats via POST /api/ping/<job>?every=24h. Queue depth, oldest-job age and failures for Celery, RQ, BullMQ, Sidekiq, Laravel.
"Node.js 20 no longer gets security fixes. Move to 24."
Lockfiles checked daily against OSV.dev, deps.dev and endoflife.date: CVEs, deprecations, end-of-life runtimes.
When something breaks, your agent gets the log lines from that moment along with the task.
POST /api/logs; search by text, level and time. Lines near an incident are attached to its task.
Counted without cookies. AI assistants are named, so you can see which ones recommend you.
No cookies; visitors are a salted hash that rotates daily. Referrers classified, with AI assistants named.
Read from Stripe, so it matches to the cent.
Stripe webhook events forwarded after the signature check; per product, with payment-to-source attribution on Pro.
Email and Slack at once, a text on Pro when customers are affected, and again when it's working.
One email thread and one Slack thread per incident; SMS on Pro when customers are affected. Recovery posts in-thread.
Every alert says what broke, since when, who is affected and what to do, always in that order. Then it gives you one line to paste into your agent.
Customers are affected right now. Email, Slack, text.
Something will become a problem if nobody fixes it. One email, no texts, no 3 a.m. anything.
Arrives in the same email thread, and the same Slack thread, once Prodbat has checked the fix from outside.
Your coding agent can see your code but not the live app, which is where real problems show up. Prodbat turns what it sees into a to-do list your agent works through, most urgent first.
Sentry and Datadog are excellent tools, and they expect someone to install SDKs, build dashboards and tune alerts. Prodbat sets itself up from one prompt and sends alerts anyone on the team can act on. Here's what changes.
| Prodbat | Sentry, Datadog and friends |
|---|---|
| Alerts written as sentences | Alerts written as stack traces and status codes |
| Set up by your agent from one prompt | Set up by you: SDKs, dashboards and alert rules |
| A to-do list your agent can work through | Graphs and traces for a person to investigate |
| The live app is checked again after each fix | You decide whether the fix worked |
| One page covering eight checks | A separate product for each kind of check |
| $19 a month, flat | Priced by events, hosts or seats |
Whichever agent built your app sets up the monitoring too. It adds a line to your pages, a few lines to your server, and a prodbat.json file listing what's watched.
Your agent gives you a link as it starts. Sign in, and monitoring is live by the time your agent finishes.
Close the laptop. If something breaks, you'll hear within 5 minutes, and your agent will already have the steps to fix it.
Claude Code, Cursor, Codex, Replit, and any agent that can run a command or connect over MCP.
Next.js, Express, Supabase Edge Functions, Cloudflare Workers, FastAPI, Django, Flask, Rails and Laravel in a few lines. Anything else with one web request.
npm, pnpm, Yarn, pip, uv, Poetry, Bundler, Composer, Go and Cargo.
Celery, RQ, BullMQ, Sidekiq and Laravel queues built in. The rest with a web request.
One subscription covers uptime, errors, background jobs, security updates and analytics, which would cost about $45 a month as separate tools. Prodbat is open source, so you can also run it on your own server for free.
Monitoring means software is watching your app while you're not. Without it, you find out your checkout broke when a customer emails you, if they bother. If people use your app or pay you, you need it.
Yes, it was written for you. Your agent installs it, alerts are written in plain English, and every alert ends with a line you can paste into your agent to get it fixed.
The browser script is 3.7 KB and loads after your page does. Your server sends reports in the background. If Prodbat is ever unreachable, your app carries on as normal.
Prodbat's instructions are fixed text, filled in only with checked values like a page address or a version number. Anything written by someone else, like an error message or a log line, is marked as evidence, and your agent is told not to follow it.
A script line in your pages, a few lines in your server code, and a prodbat.json file that lists your checkout and sign-in pages. Ask your agent to walk you through it before you deploy.
Prodbat sets no cookies and counts visitors with a code that resets every day. Whether that exempts you depends on where your visitors are. We describe the method and leave the legal call to you.
No. Your agent adds one line to the payment code you already have, which forwards each event after Stripe's signature is checked.
Only the names and versions of the packages you use, and the errors and log lines your app chooses to send.
Yes. It's open source, so you can move to your own server at any time. Removing it is a few lines your agent can take out.
Bats work nights, hear trouble before they see it, and eat bugs. We didn't need a fourth reason.
We'll load your home page from outside, check its certificate and domain renewal, and look for obvious holes like exposed config files, missing security headers and outdated frameworks. Free, and no account needed.
Hosted sign-ups open in batches. Leave your email and you'll get one message when your spot opens, and one short update a month until then.
Open source. Free to run yourself, $19 a month when we run it.
Install Prodbat on this project. Run npx prodbat init, do what it prints, deploy, then run npx prodbat verify.
Same alerts, fewer adjectives.