Your users snag the bug.
Our agent ships the fix.
Every fix arrives as a pull request for you to approve — the agent never touches your live site. Here's exactly what happens between "someone hit a bug" and "there's a PR waiting."
Three steps. One approval. Yours.
Someone snags the bug
A user hits the Snag button. We capture what actually happened — a replay of their session, the console errors, the failed requests. No "can't reproduce" ever again.
The agent works the case
It reads the evidence, finds the broken code in a private disposable copy of your repo, writes the fix, and runs your tests against it. If the tests fail, it keeps working or backs off.
You approve the PR
The fix shows up as a normal pull request — the diff, the reasoning, the evidence it used. Merge it, edit it, or reject it. Nothing ships without a human saying so.
The right tool for the job, at every step.
Not every bug deserves the same firepower. A quick check keeps the easy ones cheap; the hard ones get a specialist. Watch a bug travel through:
1 · The quick look
A fast, inexpensive check reads the evidence and decides: is this actually fixable in code? Duplicates, third-party outages, and user error get filtered out — for free.
2 · The fix
Claude reads the replay, the errors, and your code — in a private, disposable copy — then writes the fix and runs your own tests against it.
3 · The specialist
When a bug is gnarly — severe, murky, or nowhere obvious in the code — the pipeline calls in our most capable model instead of settling for a guess.
4 · The second opinion
Before you ever see a PR, a critic reviews the fix: does it match the evidence? Do the tests pass? Weak fixes get revised or rejected — not shipped.
If it can't prove the fix, it won't fake one.
No evidence in your code that matches the bug? The agent refuses to guess — it hands the ticket to a human with everything it found, instead of shipping a plausible-looking PR that wastes your review time. And because you're paying for work products, not attempts: a run that ends without a PR costs half, and your 5 free trial fixes only count when they actually produce one.
Built to be trusted with your repo.
Works on a disposable copy — never your live code. The copy is destroyed after every run.
Your own tests must pass before a PR opens. It checks its work.
A human always merges. Snag never auto-merges, ever.
Your code never trains a model. It learns from your merged fixes by lookup, not training.
It works in your repo's language.
The agent detects your stack, spins up a matching sandbox, and runs your own tests to check its fix before it opens a PR.
Things you should ask before letting an AI near your code.
What if it can't fix my bug?
Then it says so. The ticket lands with your team carrying everything the agent found — the evidence, where it looked, why it stopped. That run costs half a fix-run (trial fixes: nothing).
Can it break my site?
No. It never touches your live site or your main branch — it works on a disposable copy and the only output is a pull request that you review and merge (or don't).
Do I need to write tests for it to work?
No — it fixes bugs either way. If you do have tests, the agent runs them and a fix must pass before a PR opens, which makes its work easier to trust at a glance.
What exactly is a fix-run?
One full attempt at a bug: reading the captured evidence, analyzing your code in the sandbox, writing the fix, testing it, and opening the PR. Plans include a monthly pool; extra runs are $5 each.
What happens to my code after a run?
The sandbox copy is destroyed. We keep the bug evidence and the PR history for your workspace — never your source, and nothing is ever used to train models.
Can I use my own AI tools instead?
Yes — included from the Capture plan up, and free for life for founding workspaces. Snag exposes the same captured evidence over MCP, so Cursor, Claude Code, or Copilot can read it and open the PR themselves. The tool reference lives at snag.ac/mcp.
Your first 5 fixes are on us.
Judge the PRs like you'd review any teammate's. A run that doesn't end in a PR costs half — trial and referral runs come back in full. Monthly only — cancel anytime, prorated automatically. Capture + MCP is $29/mo flat — free for life for the founding 50 workspaces.