How it works
Label an issue. Get a reviewed PR back.
Zapat takes your backlog issues through a full pipeline — triage, implement, test, security review, code review, rework — and delivers production-ready pull requests for your approval.
The Pipeline
Six stages, start to finish.
This is not a black box. Every stage is observable in your dashboard, with narrative progress that tells you what is happening, not just that something is happening.
Label an issue or mention @zapat.
Add a `zapat:implement` label to any GitHub issue — or just comment `@zapat` to kick things off. A webhook fires within seconds. No CLI. No context switching. No prompt engineering.
Labels give you precision: `zapat:review` for code review, `zapat:test` for test coverage. Mentions give you speed: just tag `@zapat` in any issue or PR comment.
A triage agent reads the issue and your codebase.
The triage agent reads the issue title, body, labels, linked issues, and comments. Then it clones the repository and walks the codebase to understand the current implementation.
The triage agent produces a written implementation plan: which files need to change, what the approach should be, what edge cases to watch, and which specialists to recruit. This plan is visible in your dashboard.
The right specialists for each issue.
The triage agent dynamically selects specialists based on the issue — backend engineer, security reviewer, test runner, frontend engineer, and more. A bug fix gets 3 agents. A complex feature gets 5.
A simple bug fix routes to a backend engineer and a test runner. An auth issue pulls in the security reviewer and code quality reviewer. The team composition is determined per-issue, per-codebase.
A builder agent codes and tests in an isolated cloud container.
The builder agent clones your repository into a fresh, isolated ARM64 container (2 vCPU, 4 GB RAM, 21 GB ephemeral storage). It reads the codebase, follows the PM's plan, writes the changes, writes new tests, and runs the full test suite.
Each job gets a maximum of 60 minutes. The agent reads your existing code patterns, conventions, and test structure — then writes tests that match. If the implementation hits a blocker, it documents why and surfaces it to you rather than guessing.
Independent agents challenge each other.
After implementation, a review team runs in parallel. The security engineer audits for vulnerabilities. The test runner verifies coverage. The code reviewer checks quality and conventions. Each leaves structured comments on the PR.
Reviewers are intentionally independent — they do not see each other's verdicts before forming their own. This surfaces real disagreements. If security flags something the builder missed, that conflict is visible to you with full reasoning.
A clean PR, ready for your human review.
You receive a pull request with passing CI, test coverage, inline review comments from each agent, and a cost breakdown. The branch is named, the PR description explains what changed and why, and the audit trail is complete.
Auto-merge is opt-in. By default, you review and merge. For teams that want to go faster, you can configure auto-merge rules — for example, auto-merge if all agent reviews pass and CI is green.
What makes it different
Not a copilot. A pipeline.
The difference between an AI assistant and backlog automation is ownership. Zapat takes the issue and owns it end-to-end — triage through merge.
Real review, not just generation.
Most AI coding tools generate code. Zapat generates AND reviews with independent agents that challenge each other. The security engineer sees the same PR you do — and it pushes back.
Full codebase context, always.
Every agent gets your complete repository cloned into its container — not a snippet, not a summary. It understands your patterns, your naming conventions, your dependencies, and your existing tests.
Pay per issue. You control the budget.
No monthly commitment. No per-seat fee. Most issues cost $3–12. Start with $15 in free credits. Refer a friend — you both get $5, up to $35 total. Zero cost when idle.
FAQ
Common questions.
What languages and frameworks does Zapat support?
Any language you can run in a Linux container. Zapat has no hardcoded opinions about your stack — the agent reads your existing code, your package manager, and your test runner, and works with them. Common stacks (TypeScript, Python, Go, Ruby, Java) have first-class support from the agent personas.
Does Zapat have access to my code permanently?
No. Zapat clones your repository for each job, runs in an isolated container, and the container is destroyed when the job completes. Your code is never stored by Zapat. Access is scoped via GitHub App installation, which you can revoke at any time.
What happens if the agent gets it wrong?
The PR is opened for your review — you are never forced to merge. If the implementation is wrong, you close the PR and leave feedback in the issue. The audit trail tells you exactly what the agent tried and why. You are always in control of what enters your main branch.
Can I use my own AI API keys?
Yes. Bring Your Own Compute (BYOC) is available at all tiers. Connect your own Anthropic, OpenAI, or Bedrock API key and costs are deducted from your own account — not from your Zapat balance. BYOC usage does not count toward your per-issue cap.
How do cost controls work?
Tasks start as low as $3 — more complex work can cost up to $20. Every PR includes a full cost breakdown so you can see exactly what was spent. If a job approaches your limit, Zapat wraps up and opens a partial PR with notes on what remains. No runaway jobs, no surprises.
Ready to try it?
Your backlog isn't going to fix itself.
Start with $15 in free credits. No credit card required. Label an issue and get a reviewed PR back in minutes.
$15 in compute credits + you both get $5/referral. Per-issue budget controls. Cancel anytime.