BUILT FOR APPS YOU SHIPPED WITH AI

Your app broke.
Kopai already knows why.

Today it hands your AI agent the evidence to fix what broke.
Soon it will do the whole job on its own.

Connect your app free

Free while in alpha · No credit card · Your agent does the setup

checkout-app — claude

Claude Code v2.1.217

Opus 4.8 (1M context) with high effort · Claude Max

~/checkout-app

❯why can't users submit orders?
✳Crystallizing… (60s · esc to interrupt)
⏺Skill(root-cause-analysis)
⎿Successfully loaded skill
⏺Let me search for order-related errors in Kopai.
⏺Bash(npx @kopai/cli traces search --status-code ERROR --limit 20)
⎿… +116 lines (ctrl+o to expand)
⏺Found it. Let me get the full trace for more detail.
⏺Bash(npx @kopai/cli traces get 7dc92d6195b458cb597a479535049155)
⎿… +438 lines (ctrl+o to expand)
⏺Here's the root cause:

Orders fail because ProductCatalogService is unreachable.

Trace: 7dc92d6195b458cb597a479535049155

Error chain:

1. frontend → POST /api/checkout → 500

2. frontend → checkout.PlaceOrder → gRPC 13 (INTERNAL)

3. checkout → productcatalog.GetProduct → gRPC 14 (UNAVAILABLE)

▎name resolver error: produced zero addresses

Root cause: the checkout service can't resolve the DNS name for ProductCatalogService, so every order dies while it prepares the cart items.

Fix: the productcatalog container isn't running. Restart it and orders go through.

✻Cooked for 60s
❯▌▌
▶▶ auto mode on (shift+tab to cycle)● high · /effort

Your agent using Kopai on a live production bug

Set up Kopai in this app and send a test trace

Paste it into Codex, Claude Code, or any agent

WORKS WITH THE TOOLS THAT BUILT YOUR APP

Cursor
Claude
Codex
Copilot
Windsurf
Lovable
Replit
FROM THE MAKER

"Observability is no longer someone else's problem. With these tools, and the ones we ship next, you're empowered to own it."

Vladimir Adamić, maker of Kopai
Vladimir AdamićMaker of Kopai
Kopai started as an essay:Observability, Refactored
THE PROBLEM

Right now, your agent debugs blind

checkout-app — claude

❯checkout's broken. users hit pay and nothing happens9:47 PM
⏺Read(pay.ts · checkout.tsx · api/charge.ts)
⎿3 files read · the code looks fine
⏺Bash(vercel logs --prod --since 1h)
⎿Error: no access to production logs
⏺Could be a CORS issue. Or stale cache. Try redeploying?
❯redeployed. still broken10:24 PM
⏺Read(pay.ts · checkout.tsx · api/charge.ts)
⎿the same 3 files, read again
⏺Odd. Maybe clear the cache? Or check your env vars?
❯STILL broken. users are leaving11:38 PM
⏺Hmm. Could you paste the error one more time?
❯
▶▶ auto mode on (shift+tab to cycle)● high · /effort

That's not debugging, that's interrogating a witness who wasn't there.

checkout-app — claude

❯why is /checkout slow?
⏺Bash(npx @kopai/cli traces search --route /checkout)
⎿… +64 lines (ctrl+o to expand)
⏺Stripe calls time out for 4% of users. p95 is 4.2s.
⏺Cause: missing retry + timeout guard.
❯
▶▶ auto mode on (shift+tab to cycle)● high · /effort
Checkout health

Checkout is slow because Stripe calls are timing out for 4% of users.

p95 latency · /checkout▲ 4.2s
15:0015:3016:00now
THE SOLUTION

Give your agent eyes on production

Kopai watches every request, error, and slowdown in your live app, and serves it to your agent in the form agents actually use.

Most tools bolt agent access onto human dashboards. Kopai is built the other way around. You read the answer. You approve your agent's fix.

HOW IT WORKS

Three steps. Your agent does most of them.

  1. 01

    Tell your agent to set up Kopai. It does the wiring, no dashboards to configure.

  2. 02

  3. 03

For developers: OpenTelemetry under the hood. One command works with Node, Python, Go, and 8+ more.

my-app — claude

❯add kopai to my app
⏺Bash(npx @kopai/cli connect)
⎿✓ app detected · 12 api routes instrumented
⏺Done. Kopai is live, watching production.
❯
▶▶ auto mode on (shift+tab to cycle)● high · /effort

my-app — claude

❯why are signups failing?
⏺Bash(npx @kopai/cli logs search --severity ERROR)
⎿… +73 lines (ctrl+o to expand)
⏺Welcome emails bounce: SMTP_URL still points at the old provider.
❯
▶▶ auto mode on (shift+tab to cycle)● high · /effort

my-app — claude

❯fix it
⏺Update(config/email.ts)
⎿✓ 2 lines changed · retry guard added
⏺PR opened — approve it and I'll ship.
❯approved, ship it
✻deployed · signups recovering
❯
▶▶ auto mode on (shift+tab to cycle)● high · /effort
WHAT'S IN THE ALPHA

Everything here works today.

No vision slides. What's live today is below, the rest is on the roadmap, clearly marked.

your agent · terminal
[main]$:why did checkout fail?
›asking kopai · traces + logs
✗stripe gateway timeout · 23 hits
●failing since 2:14am · live
POST /api/checkout → 504 · stripe.charges.create timed out after 30s
›fix: retry with 8s timeout
✓shipped · resolved
[main]$:▌

Debugging with evidence

Your agent pulls up what your app actually did, then fixes it with evidence instead of guesses. Claude works with Kopai today.

show me failed signups this weekAsk
Failed signups · last 7 dayslive
142failed this week
▲ 18%
37
MTWTFSS

Dashboards on demand

Say “show me failed signups this week” and the live view builds itself. No query language, no setup.

Open source. Your data stays yours.

The code is public on GitHub. Your data stays encrypted, isolated, and portable: take it anywhere, any time.

View on GitHub  →

Want raw charts and traces? One click deeper. Don't? You'll never see them.

THE OBVIOUS QUESTION

“Isn’t this just what Sentry does?”

Fair question. (Never heard of Sentry? Skip ahead. You’re exactly who we built this for.)

Every monitoring tool was built for an engineer who reads dashboards for a living. Sentry’s AI debugger is paywalled per contributor, on top of a paid plan.

Name one built for someone who ships with an AI agent. Exactly.

EVERY OTHER TOOL

Dashboards for humans

  • You read the charts and traces yourself.
  • AI debugging is a per-seat add-on.
KOPAI

Answers for your agent

  • Your agent asks, gets evidence, and ships the fix.
  • AI debugging is the whole product.

Love dashboards? Sentry’s good at them. Want answers? That’s Kopai.

ROADMAP TO V1

Toward an app that fixes itself

NOW

Answers when you ask

Connect in minutes, debug with evidence, and build dashboards on demand. All live today.

NEXT - Q3 26'

Alerting

Kopai tells you something's wrong before your users do. Not in the alpha yet.

LATER - Q4 26'

Autonomous excavations

Kopai spots trouble on its own, investigates, and hands you a fix to approve. Nothing ships without your OK.

Alpha users steer this. Tell us what hurts.

STAY IN THE LOOP

Not ready to connect an app?

Leave your email and watch Kopai grow up. One short note when something new ships.

No spam. Unsubscribe anytime.

Before you ask

I'm not a real developer. Is this for me?

You're exactly who it's for. If you shipped something real with Lovable, Bolt, v0, Cursor, or Claude, you qualify, and you'll never be asked to read a log.

Is this another thing I have to maintain?

No. Your agent connects it in minutes, and after that there's nothing to configure and no thresholds to tune.

What does it cost?

Nothing while it's in alpha, no credit card. After that: one simple price that grows with your app, never per seat.

What happens to my app's data?

Kopai collects what your app reports about production: errors, requests, and performance signals. It's stored on open standards, and you can take it with you anytime.

Is this one of those AI tools that runs my servers for me?

No. Kopai watches and explains; changes to your app only ship when you approve them.

How rough is “alpha”, honestly?

Real enough to connect an app and debug a live bug with today, rough in places everywhere else. Early users get a say in what gets fixed first.

Will it work with my app?

If your app has a backend, almost certainly. Kopai speaks OpenTelemetry, which covers every major language, and your agent handles the wiring.

FREE WHILE IN ALPHA

Let your agent see what it built

The next 2 a.m. bug gets an answer, not a guessing session.

Connect your app free

Free while in alpha  ·  No credit card  ·  Your agent does the setup