// QA that runs itself

Stop clicking through
your app before every release.

Describe a flow once. Traceback runs it on every deploy, catches what breaks, and proposes a fix when something breaks.

436 developers on the waitlist · Private beta · March 2025

01 / goal
goal

“Make sure a new user can sign up and reach the dashboard.”

02 / run
tb_8x4k2m23.4s
3 passed1 failed2 skipped
03 / fix
ConfirmEmail.tsx:47

Button renders as <span> — check conditional render path.

Push to GitHubPush to Linear
01 /The Problem

Writing code stopped
being the hard part.

Copilot, Cursor, Claude Code — engineers using AI tools ship significantly more code than they did two years ago. The bottleneck shifted. Now the hard part is verifying that code actually works.

01AI coding tools ship code faster. More features per sprint, more PRs per week, more surface area for things to break.
02QA didn't keep up. Someone still has to click through the app before every release. That someone is usually the developer who wrote the feature.
03The gap is where bugs live. The faster you ship without a testing layer, the more often users find things before you do.
Traceback is the testing layer. Describe the flows that need to work, and it runs them automatically — before every deploy, on every PR, on a schedule.
02 /How It Works

Describe it once.
Forget it forever.

01 / Define

Write what should work.

One sentence. No selectors, no framework to install, no scripts to write. Plain English is the entire test definition.

goal

“Make sure a new user can sign up and reach the dashboard.”

02 / Execute

A real browser tests it.

Chromium opens. The agent navigates, clicks, fills forms exactly like a user. Under 90 seconds per run.

navigate to /signup
fill email field
click Create Account
verify email button
03 / Report

Get the fix, not the error.

Not “element not found.” A plain-English description of what broke, why, and what to change. Push to GitHub or Linear in one click.

ConfirmEmail.tsx:47

Button renders as <span>. Check conditional render path.

03 /Your Stack

Already in your workflow.

GitHub

Runs on every PR against your preview deploy. Results post as a status check. Merge is blocked if tests fail.

Linear

Push failure reports directly to Linear as tickets in one click. Pre-populated with context from the run.

Slack

Failure alerts fire immediately with what broke. Use @traceback in any channel to trigger runs naturally.

Notion

Sync run results and failure reports to your team's Notion workspace automatically.

Claude

Give your Claude Code agent a testing layer. It runs tests, reads reports, and fixes bugs autonomously.

MCP

Five tools — create_test, list_tests, run_test, get_report, debug_failure — for any MCP-compatible agent.

04 /Two Modes

Two modes. Zero test code.

Autonomous Mode

Write a goal. We figure out the rest.

A browser agent navigates your app visually. No selectors, no scripts, no SDK. It finds what a real user would find: broken flows, missing elements, logic failures.

Input“Make sure checkout works”
AgentNavigates visually
ResultFAILED · step 04
No selectorsVision-basedAny web app

Deterministic Mode

Exact steps. Self-healing selectors.

Write each step in plain English. Traceback generates the Playwright. When your UI changes and a selector breaks, a vision model finds the element, repairs it, and continues. No false failures. No manual fixes.

01navigate /loginPASSED
02fill email fieldPASSED
03click Sign in buttonHEALED
04assert Dashboard loadsPASSED
HEALEDselector broke. Vision model found the element and repaired it.
05 /The Output

The final output is the fix.

Not just pass/fail. A full breakdown pointing at the exact cause in your code.

RUN IDtb_8x4k2m
TESTcheckout_flow_v2
STATUSFAILED
DURATION23.4s
TRIGGERED BYgithub / pr #241
Steps
01navigate to /signup
02fill email field
03fill password field
04click verify email button
05assert dashboard visible
06assert welcome message
Viewports
DesktopFAILED
MobilePASSED
TabletPASSED
2 / 3 viewports passed
Run Failed at Step 5
00:18TimeoutError: Waiting for selector [data-testid="verify-email-btn"] — waited 5000ms
00:18at BrowserAgent.waitForSelector (agent/browser.ts:142)
00:18at Step.execute (agent/step.ts:89)
00:18at TestRunner.run (runner/index.ts:201)
Total Steps14
Assertions Passed3
Credits Used12
Identified Bug

The “Verify Email” button is not interactable on the desktop viewport. It renders as a <span> instead of a <button>, preventing click events from firing. Likely caused by a conditional render path in ConfirmEmail.tsx when the verification token is absent from state.

View in Issues →
06 /What Teams Say

“I was spending 20 minutes manually testing checkout every time we shipped. Offloading that to an agent that just pings Slack when it fails has given me my sanity back.”

“Our old E2E tests were so brittle we eventually just stopped running them. The fact that Traceback can survive basic UI changes and fix its own selectors is exactly what we needed.”

“Getting a vague 'timeout' error from CI used to ruin my afternoon. Traceback actually points out the specific component that broke. It cuts my debugging time way down.”

07 /Who It's For

Built for teams shipping fast.

Small teams

No QA team. No problem.

Traceback becomes your QA function. Describe the flows that matter and they run on every deploy. You stop being the last line of defense.

GitHub workflow

Never break main again.

Traceback runs on every PR against your preview deployment. Results post as a status check. If checkout breaks, the merge is blocked.

Production monitoring

Know before your users do.

Smoke tests run against production on a schedule. When a test fails, a Slack alert fires with what broke and a link to the full report.

AI coding agents

AI agents that test their own work.

The agent calls Traceback's MCP server, reads the failure report, applies the fix before merging. No human in the loop.

Critical user flows

Signup and checkout work. Every time.

Autonomous mode navigates these flows end-to-end with a real browser. Catches broken email verification, payment form errors, redirect loops.

Refactoring safely

Refactor confidently. Tests adapt.

Self-healing selectors adapt to UI changes automatically. You see what changed without a sea of red failures.

08 /FAQ
Do I need to write any test code?
No. You write a goal in plain English. Traceback generates and runs everything. In deterministic mode, you write steps in plain English and Traceback generates the Playwright under the hood.
What happens when my UI changes and selectors break?
In deterministic mode, a vision model finds the element visually and repairs the selector automatically. You see a HEALED status instead of a failure. No manual fix required.
How is this different from Playwright or Cypress?
Those tools require you to write and maintain test code. When selectors break, you fix them. Traceback writes no test code, runs tests autonomously, and self-heals when your UI changes.
Can it test apps that require login?
Yes. You can set up auth credentials and session state as part of the test configuration. Traceback handles login flows before running the rest of the test.
How does the GitHub integration work?
Connect your repo and Traceback runs against your preview deployment on every PR. Results post as a commit status check. A failing test blocks the merge until it's fixed.
Is the beta actually open?
Yes. You can sign up and run your first test today. Book a demo if you want a walkthrough of your specific use case first.
09 /Pricing

Start free. Test as you ship.

One credit = one agent action (click, fill, navigate). Overages at $0.05 / credit.

PlanPriceCredits / moSeats
Free$01001
Starter$29 / seat600max 2
Promost popular$99 / seat2,000max 3
Team$199 / seat4,000unlimited
EnterpriseCustomCustomunlimited
Book a Demo

One test run uses 8 to 25 credits. Most teams run hundreds of tests a month on Pro without thinking about cost.

Your next release
ships clean.

Or Traceback tells you exactly why it didn't.

Free to start · No credit card required · Beta now open