GitAuto
GitAuto is a set-and-forget GitHub App for dev teams: it writes unit tests, opens pull requests, runs the CI, fixes failing checks and merges. It targets 80-90% test coverage and bills per pull request, from 2 USD.
What is GitAuto?
GitAuto is a set-and-forget coding agent that opens unit test pull requests on GitHub. It is published by GitAuto, Inc., a US company founded on 10 May 2024, incorporated in Delaware and based in San Francisco, led by founder and CEO Wes Nishio. The product ships as a GitHub App plus a web dashboard: there is no mobile app and no browser extension.
The problem it targets is the one every team recognizes and nobody prioritizes, a codebase whose coverage sits far below where it should be and nobody with the time to fix it. The home page frames the promise as going from 0% to 90% test coverage, while the site's llms.txt states 80%+. The vendor claims more than 4,000 repositories, a 99.5% test pass rate and a 94.9% PR merge rate.
The mechanism is a nine-step loop, described on the product page and repeated every day: detect, learn, open, write, run, fix, address, sync, merge. GitAuto reads the coverage reports your CI publishes, learns the testing conventions already present in the repository, opens one pull request per target file, runs the CI, fixes the failures, answers review comments, re-syncs the branch and merges. Four triggers fire it: a schedule of up to 24 runs a day, a failing test, a review comment, or a manual pick from the dashboard. What it learns from reviews and CI failures is persisted as reusable rules in a GITAUTO.md file versioned in your repository, alongside repository rules that you configure in the web interface and that GitAuto never edits.
Four models are selectable, and the model sets the price per PR: Google Gemma 4 31B at 2 USD, Claude Sonnet 4.6 or Google Gemini 2.5 Flash at 4 USD, Claude Opus 4.8 at 8 USD, with context windows from 100K to 1M tokens. Python, JavaScript, TypeScript, Java, Go, PHP, Ruby and Flutter are documented, and the FAQ extends support to any language that has a test runner, up to COBOL and Fortran. Output language is configurable per repository across 74 languages, though PR titles and bodies stay in English. Installation is one click from the GitHub Marketplace, claimed at under two minutes. The hard limit is the platform: GitHub only.
What it does
- Writes and updates unit tests that follow the conventions already used in the repository
- Opens one pull request per target file, on a schedule you set
- Runs the CI, reads the logs and fixes failing checks until everything is green
- Merges automatically once every required check passes
- Detects uncovered files from the coverage reports produced by your CI
- Answers review comments and re-syncs the branch when other pull requests are merged
- Charts coverage progress file by file in a web dashboard
When to use GitAuto / When not to
A quick filter to help you decide if GitAuto is the right fit.
When to use GitAuto
- Engineering teams sitting on a large legacy codebase with no real test culture, who need coverage to climb without pausing feature work.
- Teams held to a high coverage target with thin QA capacity, such as one QA engineer for fifteen developers.
- Teams working against a contractual quality deadline, where a coverage report has to be produced for a client or an auditor.
- Regulated and safety-critical codebases: the setup documentation points to 90%+ for fintech and healthcare, and 95%+ for aerospace, automotive and medical devices.
- Teams already on GitHub whose CI publishes coverage reports, since GitAuto drives everything from those artifacts.
When not to use GitAuto
- Teams hosting on GitLab or Bitbucket: GitAuto is a GitHub App only, and the FAQ offers no workaround.
- Repositories with no test framework or test runner in place, since support is explicitly conditional on a runner existing.
- Teams whose CI produces no coverage report, because the coverage-driven trigger then has nothing to read.
- Anyone looking for feature code, refactoring or documentation: GitAuto writes test code and nothing else.
- Organizations that require a signed DPA, formal GDPR guarantees or hosting outside the United States, none of which is published or offered.
How to use GitAuto
A typical end-to-end flow, from setup to results.
- Install the GitAuto GitHub App from the GitHub Marketplace: one click, under two minutes, no credit card. An organization owner or repository admin is required, and anyone else can send an approval request.
- Select the repositories GitAuto may work on, either all of them or a shortlist.
- Wire up coverage reporting. On a single repository GitAuto detects the language and opens its own 'Set up test coverage workflow' pull request; otherwise follow the per-language guides.
- Read the coverage charts and set a target: 80% as a standard, 90%+ for regulated codebases, 95%+ for safety-critical systems.
- Configure repository rules from the dashboard: code style, test naming conventions, comments, test constants.
- Enable the scheduled trigger and size its frequency. The worked example given: 500 files below target over three months, roughly 60 working days, comes out at 8 to 10 runs a day.
- Review and merge the pull requests. Your review comments feed GITAUTO.md and reduce corrections on later runs.
- Buy credits when you want to scale up, for more pull requests per day and more repositories.
Pros & Cons
Pros
- The loop is genuinely end-to-end: not only writing tests, but running the CI, fixing failures, answering reviews, re-syncing and merging.
- Set-and-forget in practice. Once scheduling is on, there is no prompting and no manual step.
- No subscription and no commitment: you pay per pull request, with a 24 USD welcome credit per GitHub organization and no credit card to start.
- You choose the model, and therefore the cost/quality trade-off, from 2 to 8 USD per pull request.
- Learning is persistent, versioned and visible: GITAUTO.md lives in your repository and your team can edit it.
- Language-agnostic, covering anything with a test runner, with output localizable across 74 languages.
- A clear stated position on data (no code retained, no training, GitHub secrets excluded), backed by unusually detailed documentation on internals, guardrails and ROI methodology.
Cons
- GitHub only: no GitLab, no Bitbucket, no workaround.
- Deliberately narrow scope. It writes unit tests, not feature code.
- It depends entirely on the quality of the coverage reports your own CI produces.
- Costs are variable and hard to budget in advance, following the number of pull requests and the model you pick, and credits expire after one month on Free and one year on paid tiers.
- The free tier is very tight: one repository, one user, one coverage upload, one scheduled run per day, default branch only.
- Nothing on GDPR, no DPA, no EU representative, no published sub-processor list and no security certification (no SOC 2, no ISO 27001), which makes the vendor hard to clear for a European buyer.
- US-only hosting with no region choice, a single email address for legal, privacy and support, paid plans prioritized for support, no mobile app, no public API, and a full FAQ that only renders with JavaScript enabled.
Pricing & Plans
There is a permanent free plan and no subscription of any kind: GitAuto is billed per pull request from prepaid credits, and the lowest price point is 2.00 USD per pull request with Google Gemma 4 31B. The other tiers are 4.00 USD (Claude Sonnet 4.6 or Google Gemini 2.5 Flash) and 8.00 USD (Claude Opus 4.8). Each GitHub organization receives a 24 USD welcome credit, equivalent to twelve pull requests at the entry rate or three at the top rate, and no credit card is required to begin. Credit purchases start at 10 USD, approximately one pull request at the highest rate, and are capped at 5,000 USD per transaction, above which the sales contact takes over. Credits expire after one month on the free tier and after one year on the paid tiers. Optional auto-reload triggers below a default 10 USD balance, and a customizable monthly spending cap defaults to 5,000 USD. When the balance reaches 0 USD, pull request generation stops.
- Google Gemma 4 31B
- 100K context
- 16K output
- 1 repository
- 1 user
- 1 coverage upload
- 1 reference URL and 1 reference file
- default branch only
- Google Gemma 4 31B
- 250K context
- 32K output
- unlimited repositories and users
- unlimited coverage uploads
- any base branch
- scheduling up to 24 times a day
- credits valid for 1 year.
- Claude Sonnet 4.6 or Google Gemini 2.5 Flash
- 500K context
- 64K output
- with the same unlimited allowances as the 2 USD tier.
- Claude Opus 4.8
- 1M context
- 128K output
- the same unlimited allowances
- flagged 'Best Quality'.
- GITAUTO.md
- repository rules
- the four triggers
- unlimited retries on failing tests
- output in 70+ languages
- Self LLM API Key
- a 10 USD minimum purchase
- and customizable auto-reload and spending caps.
- Custom pricing through the sales contact
- required above the 5
- 000 USD per-transaction ceiling.
Data, GDPR & hosting
A consolidated view of how GitAuto handles your data.
GDPR overview
There is no mention of the GDPR anywhere on the site. Across the thirteen pages reviewed and their raw HTML, GDPR, General Data Protection Regulation, EEA and Article 27 return zero occurrences. No EU representative is designated and no data processing agreement is published or offered. The privacy policy, revised on 16 July 2024, addresses United States law only: the CCPA (California), the CTDPA (Connecticut), the VCDPA (Virginia) and a generic section for US residents. Those laws grant access, correction, deletion, portability, objection to profiling and targeted advertising, and withdrawal of consent, with appeal windows of 45 days under the CCPA and 60 days under the CTDPA and VCDPA. The Terms warn that the Services are hosted in the United States. GitAuto, Inc. declares no EU establishment, so a European user has no published basis on which to assess transfers.
Who owns the data?
GitAuto's Terms are explicit: you retain full ownership of your Contributions, and the company asserts no ownership over the code you submit or the tests it writes in your repository. Nothing is licensed away. On storage, the vendor states that source code does not remain on its servers; when a run fails, only error logs containing no code are kept on AWS Lambda for debugging. Payment details are held by Stripe, not by GitAuto. Under the privacy policy revised on 16 July 2024, personal data is shared with third parties only in a corporate transaction such as a merger, acquisition or financing, and none has been sold or shared in the past twelve months.
Reuse rights
Because ownership stays with you, the tests GitAuto commits are yours to keep, modify, publish or ship without asking permission. What travels the other way is tightly scoped. Installing the app alone sends nothing: transmission begins only when a pull request is created, at which point the repository file tree is sent and only the files the context actually requires are referenced. The inputs are the PR title and body, review comments, the file tree, the contents of linked files, your repository rules and the GITAUTO.md file. GitHub secrets are excluded. Generation runs through Anthropic's Claude API, which the vendor states does not train on or retain customer data, so code is processed then discarded. An email address is collected at installation only if it is public on your GitHub profile, and is used for account management, notifications and outreach, with an opt-out available. The privacy policy revised on 16 July 2024 states that no personal data has been sold or shared in the preceding twelve months.
Data retention & training
Hosting summary
The Terms of Service state that the Services are hosted in the United States, with an explicit warning to users accessing them from other regions. That is the only jurisdictional statement published: no hosting region is named, no data center is identified, and no region choice is offered. Several third parties are named across the FAQ and the privacy policy without any formal sub-processor list ever being published. Source code is processed by Anthropic's Claude API, while Google Gemini and Gemma models are offered as alternatives. Error logs, which contain no code, are retained on AWS Lambda. Payment data is stored by Stripe. On the infrastructure side the domain resolves to 64.29.17.65, a United States address in AS16509 (Amazon), but this is an anycast CDN node and proves nothing about where customer data actually sits. For a European organization the practical reading is straightforward: processing happens in the United States, under US law, with no published transfer mechanism.
Things to keep in mind
Risks and trade-offs to weigh before adopting GitAuto.
- Platform lock-in: everything is tied to GitHub, so a move to GitLab or Bitbucket ends the relationship with no migration path.
- Cost scales with volume. Scheduling up to twenty-four runs a day without deliberately setting the spend cap, which defaults to 5,000 USD a month, can get expensive fast, and credits expire after one month on Free and one year on paid tiers.
- Auto-merge is still auto-merge. GitAuto only merges when the CI passes and nothing but test files changed, but code lands in your repository with no human in the loop.
- Your source code is transmitted to third parties (Anthropic, Google) at generation time. The vendor states it is neither retained nor used for training, but it does leave your perimeter.
- No formal sub-processor list, no DPA, no GDPR mention and no published security certification, with hosting in the United States only: that compliance gap is carried by you, not by the vendor.
- Repository rules are yours to maintain, and a badly written rule propagates silently to every generated pull request.
- Coverage is a proxy, not a guarantee. A team that treats the percentage as the objective can end up with a green dashboard and tests that assert very little, and reviewers who stop reading generated pull requests closely lose exactly the judgement the tool was meant to free up.
Setup & Integrations
Technical difficulty
Low for the app itself, moderate for the prerequisite. Installing the GitHub App is one click from the Marketplace, claimed at under two minutes, with no credit card, though you need organization owner or repository admin rights. The real requirement sits upstream: a CI workflow that runs your tests and publishes a coverage report as a GitHub Actions artifact. On a single repository GitAuto detects the language and opens that configuration PR itself; otherwise you follow per-language guides. Then set repository rules, a coverage target and the schedule. The vendor offers hands-on help for non-standard CI or permission blocks.
Deployment
Integrations
Supported languages
Behind GitAuto
Social
Resources
All the official URLs gathered for verification and reference.
Alternatives
Tools that compete with or complement GitAuto.
Frequently asked questions
Does GitAuto work anywhere other than GitHub?
What is actually sent to the AI?
Is my code stored anywhere, or used to train models?
Which programming languages are supported?
Can GitAuto write in a language other than English?
Who can install GitAuto on a repository?
What is the minimum purchase, and what happens when credits run out?
Are pull requests merged automatically?
Is my email address collected?
What support is available?
Should you pick GitAuto?
GitAuto does one thing: it drives unit test coverage upward on GitHub repositories. Nothing else. That narrowness reads as a deliberate choice rather than an oversight, and the product is better focused for it.
What sets it apart is that the automation does not stop at generating tests. The nine-step loop runs the CI, reads the failures, fixes them, answers review comments, re-syncs the branch and merges, on a schedule you set and without prompting. For a team carrying a legacy codebase with no test culture, or one held to a contractual coverage figure, that is the difference between a tool and a colleague.
The pricing is legible: no subscription, no seats, a fixed amount per pull request from 2.00 USD, a 24 USD welcome credit and no credit card to start. You pick the model and therefore the cost. The flip side is that spend follows volume, so the monthly cap deserves to be set deliberately before scheduling twenty-four runs a day.
Two reservations are worth weighing. The first is platform lock-in: GitHub is mandatory, and a migration to GitLab or Bitbucket ends the relationship. The second is the compliance file, which is thin outside the United States. There is no GDPR mention anywhere on the site, no DPA, no EU representative, no published sub-processor list and no security certification, with hosting in the US only. A European buyer with a formal procurement process will struggle to clear it, however sound the vendor's stated data practices are.
GitAuto, Inc. is young: incorporated in May 2024, one named founder, a small team. Weighed against what a trial actually costs, that is a reasonable bet for a GitHub team with a coverage problem, and a difficult one for a regulated European enterprise.
- Choosing a selection results in a full page refresh.
- Opens in a new window.