Bugpilot: Console & Network Bug Reports for AI — Minus the Noise

Your AI assistant can't debug what it can't see. Bugpilot records a browser tab while you reproduce a bug, then gives you one structured Markdown report — console output with stack traces, network requests with the bodies of the calls that failed, clicks, form state, DOM and React state — ready to paste into any AI No account. No server. Nothing leaves your machine. THE PROBLEM Debugging with an AI usually starts the same way: you describe the bug from memory, paste an error you half-remember, then spend three or four replies answering questions the assistant has to ask because the evidence isn't in front of it. • What did the request return? • What was the component state when it threw? • What did you actually click before it broke? • Was there anything in the console before the error? The browser had all of that at the moment the bug happened. The problem is collecting it and presenting it in a form a model can use. That is the whole job Bugpilot does. HOW IT WORKS 1. Open the page that's broken and click Record. (Ctrl+Shift+B, or Cmd+Shift+B on Mac, opens Bugpilot.) 2. Reproduce the bug the way a user would. 3. Click Stop. The report opens in a new tab. Copy it, or download it as a file. No time limit, no event cap, no auto-stop. A recording ends when you stop it. While you record, Chrome shows its own yellow "debugging this browser" banner. That's the Chrome Debugger API doing its job — the only way in Manifest V3 to read what a failed request actually returned. It clears the moment you press Stop. WHAT GETS CAPTURED • Console log, info, warn and error with stack traces, plus uncaught exceptions and unhandled promise rejections • Network requests with method, URL, status, timing and initiator — plus full request and response bodies for calls that failed, and for successful calls that finished within two seconds before an error, labelled with a confidence level • Clicks, with the selector, the visible text, and the state of every field in the surrounding form • Form inputs, committed on blur, and both SPA and full-page navigations • The DOM at the moment of the error • The React component tree, props and useState/useReducer values at the moment it threw • WebP screenshots at record start, on console errors, and at stop — off by default; turn them on in Settings Every event is tagged with an ID (E001, E002…) and a timestamp relative to the start of the recording, so the report can cross-reference cause and effect instead of leaving the model to guess. A small rule engine then reads the capture and proposes a most-likely cause with a confidence level, printed as a TL;DR at the top. Four deterministic rules — no model call, nothing inferred, nothing sent anywhere. SMART SUMMARY — SAME EVIDENCE, SMALLER REPORT A real capture is mostly noise: dozens of successful script, stylesheet, font and image requests, and the same console line repeated forty times. That noise costs you context window every single time you paste. Smart Summary strips it, and collapses identical messages into one line with a repeat count. What it never drops, by rule: • Errors and warnings • Any request that failed — including a 404 font or a 500 chunk • Any request correlated with an error • Document, XHR and Fetch requests • Anything another event cross-references • Anything within three seconds of an error It's opt-in, off by default, and applied at format time only. The stored capture is always the full recording, so switching the summary off re-renders the complete report with nothing lost. On our own test captures it removed between 55% and 96% of the report text, depending on how noisy the capture was. A clean capture legitimately saves little. You don't have to take the range on faith — the report page shows your capture's own measured figure before you switch it on. WHO IT'S FOR • You're building on Lovable, Bolt, v0 or Replit and the app breaks. One button turns "it's broken" into a report the AI can act on. No DevTools knowledge required — the report names the failing request and the failing component for you. • You work in Cursor, Copilot, Claude or ChatGPT. You already paste context by hand, and you know exactly what the round trips cost. • You're a front-end or full-stack engineer. You can debug this yourself; what you hate is transcribing it for someone else. • You're in QA, support or product. You can file a complete bug report without asking a developer to reproduce it first. • You can't paste customer data into a third-party capture service. There is no service. PRIVACY Everything happens on your machine. Captures are recorded, redacted, stored and formatted locally, in your browser's own storage. No Bugpilot account, no Bugpilot server, no telemetry, no analytics. Passwords, API keys, bearer tokens, JWTs, credit card numbers, auth headers and sensitive URL parameters are redacted out of network traffic, form fields, URLs, DOM snapshots and React state before anything is stored. Password fields are redacted by field identity rather than by pattern, so the value never reaches storage at all. Every report ends with a summary of what was removed. One narrow exception: buying or verifying Pro goes through ExtensionPay, our payment provider, so the extension asks extensionpay.com whether this install is paid, using an anonymous identifier. Your captures are never part of that request. The report is plain text. Read it before you paste it. FREE AND PRO Free is not a trial. It covers the whole loop: unlimited recording with no time limit, every capture type above, redaction, the TL;DR and heuristic analysis, Markdown copy and download, error-screenshot copy, a .zip of report plus screenshots, browsing your 10 most recent captures, and 5 Smart Summary reports. Pro is a one-time payment — $28 for the first 200 buyers, $44 after that. Lifetime, not a subscription. It adds: • Unlimited Smart Summary • The full React component tree with props and hook state, instead of just …