$ error fires → pull request appears
~/production
Your AI bug-fixing agent. And the control room to run it.
bugzero picks up new errors from Sentry, PostHog or a custom webhook, fixes them inside your own GitHub Actions and opens a pull request for you to review. One dashboard to choose the model, version the agent's instructions and watch every fix.
└ no credit card · your code stays in your github · your own claude or gpt key
TypeError: Cannot read properties of undefined (reading 'id')
- habitat
- src/lib/user.ts:42 · acme/web
- source
- sentry · production
- status
- caught · agent dispatched
From a stack trace to a pull request — inside your own GitHub.
Watch one error travel through the loop. bugzero only ever moves metadata; the code and your AI key stay in your Actions runner.
Most tools give you an agent or a dashboard. You get both.
Does the fixing.
- Reads the stack trace, finds the root cause and writes a minimal fix
- Runs inside your own GitHub Actions — your code never leaves your infrastructure
- Uses your own Claude or GPT key, billed by your provider at cost
- Always ships a pull request — nothing merges without your review
Keeps you in charge.
- Connect Sentry, PostHog or a custom webhook and map each project to a repository
- Pick the model and effort — switch between Claude and GPT anytime
- Version the agent's instructions, compare changes, roll back in one click
- See every run — root cause, pull request, tokens and cost — plus a daily safety cap
Pay for the agent, not the tokens.
You bring your own AI key, so tokens are billed by Anthropic or OpenAI at cost — we never mark them up. Start free, go unlimited for $9.
Free
For trying bugzero or a small side project.
$0 / month
5 fixes, then wait for the 1st
- 5 fixes per month
- Every Claude and GPT model
- Sentry, PostHog and custom webhooks
- Versioned agent instructions
- No credit card required
Pro
[unlimited]For teams shipping to production.
$9 / month
your key (at cost) + $9 = ∞ fixes
- Unlimited fixes
- Everything in Free
- A daily safety cap you control
- Cancel anytime
└ a fix counts when the agent opens a pull request — failed runs are always free. AI tokens are billed by your provider at cost, never through us.
Questions, answered.
What exactly is bugzero?
Two things in one: an AI agent that fixes production errors, and the control room you run it from. The agent works inside your GitHub Actions; the dashboard is where you connect error sources, pick the model, version its instructions and review every fix.
What if the fix is wrong?
It's always a pull request — you review it and merge only if it's right, nothing ships automatically. If fixes miss your conventions, adjust the agent's instructions: every change is versioned, so you can compare versions and roll back in one click.
Is my source code safe?
Yes — bugzero's GitHub App can't read your code at all; its only permission is to start the workflow. The agent runs in your own GitHub Actions runner, so your code and your AI key never leave your infrastructure. bugzero only receives the run status, the agent's summary and the PR link.
Why is Pro only $9?
Because you bring your own AI key. The expensive part — the tokens — is billed to you by Anthropic or OpenAI at cost, so we only charge for the agent and the control room. Unlimited fixes, one flat price.
What counts as a fix?
A run that opens a pull request. Failed runs never count. Free includes 5 fixes a month; Pro is unlimited, with a daily safety cap you control so an error storm can't run up your AI bill.
Which error trackers and models are supported?
Sentry and PostHog, plus a custom webhook for anything else — any app, logger or tool that can send an HTTP request. The agent runs on the latest Claude (Opus, Sonnet, Haiku, Fable) or GPT models — switch anytime.
Does it work with private repos?
Yes. The agent runs inside your own GitHub Actions, and the bugzero GitHub App only needs permission to start workflows on the repositories you pick — your code stays in your infrastructure.
What languages are supported?
Any language — the agent reads code, it doesn't execute it.