Pie
Pie is an autonomous QA platform whose vision-based AI agents explore your web, iOS or Android app like a real user, test every pull request, and open a fix pull request once a human has confirmed the bug.
What is Pie?
Pie is an autonomous QA platform published by PieLabs, Inc. of Oakland, California, under the tagline Ship the feature. Not the bug. Rather than running scripts, it sends AI agents through the product the way a customer would, on web, native iOS and Android, and browser extensions. The technical difference is visual recognition: Pie identifies a button by how it looks and where it sits, not by a CSS class, an XPath or an element ID, so a redesign does not break coverage and tests heal themselves. Work happens in three phases the vendor calls Push, Probe and Pass. Push takes a URL, an .apk or .app build, or an extension upload, plus credentials when the app is gated. Probe runs discovery, generation and execution at once: agents map the product, group it into key features, adopt user personas to decide what a person would be trying to achieve, then spread hundreds of parallel runs across what Pie calls the Pie Farm, turning an overnight regression suite into roughly thirty minutes. Pass returns a Readiness Score with deduplicated issues, so fifty failures caused by one broken call become a single issue carrying a root cause, reproduction steps, a frame-by-frame screenshot sequence and an expected-versus-actual breakdown. Two guardrails bound the output. Every anomaly is reproduced and reviewed by a QA professional on Pie's side before it reaches you, and every fix arrives as an ordinary pull request that Pie never merges, with a comment-only mode available. Nothing is installed inside your stack: no SDK, no source code access, testing runs from the outside against staging or production. Coverage is advertised at 80% end-to-end from day one, with setup in about five minutes. Supported technologies include Swift, Kotlin, Kotlin Multiplatform, React Native, Flutter and webviews. Pie plugs into GitHub Actions, Bitbucket Pipelines and a REST API for Jenkins, CircleCI, GitLab or AWS CodePipeline, plus Jira, Slack and an MCP server for Cursor, Claude Code, Claude Desktop and VS Code. On security it holds SOC 2 Type 2, monitored continuously through Sprinto across 41 controls, with AES-256 at rest, TLS 1.2+ in transit, sandboxes destroyed after each run, and a contractual promise that customer data never trains a model.
What it does
- Explore a web or mobile app autonomously and map its real user journeys
- Generate test cases from that exploration, with no script or selector to write
- Read every pull request and exercise the change it introduces
- Catch functional and experience bugs a passing status code would hide
- Open a fix pull request with the root cause and visual evidence attached
- Create a test from a plain-English instruction in under a minute
- Cluster failures into one deduplicated issue and report it to Slack or Jira
When to use Pie / When not to
A quick filter to help you decide if Pie is the right fit.
When to use Pie
- Product and engineering teams shipping faster than their quality process can keep up with
- Mobile teams with native iOS and Android apps built in Swift, Kotlin, React Native or Flutter
- Teams with no dedicated SDET who cannot staff a scripted end-to-end suite
- Teams leaning on AI coding assistants and watching code volume outrun their test coverage
- Startups replacing a large manual regression effort, as the published Fi case study describes
When not to use Pie
- Fully staffed quality teams that are happy maintaining their own Playwright suite, as the vendor itself concedes
- Anyone looking for security or penetration testing, which Pie explicitly does not cover
- Teams wanting unit or component testing, since Pie stops at the end-to-end layer
- Engineers who need their test code to live in their own repository and stay vendor-neutral
- Small budgets needing a public price list, since Pie quotes only after scanning your app
How to use Pie
A typical end-to-end flow, from setup to results.
- Create an account on the Pie start page; no credit card is requested
- Point Pie at your web app URL, or upload the .apk or .app build, or the browser extension
- Add test credentials if the application sits behind authentication
- Let the agents explore and map your features during the Probe phase
- Connect your repository through the ready-made GitHub Action, Bitbucket Pipelines, or one REST API call from Jenkins, CircleCI, GitLab or AWS CodePipeline
- Add your own cases in plain English from the Pie assistant, or import an existing suite as CSV or Cucumber/Gherkin
- Wire Slack for notifications and Jira for automatic ticket creation
- Install the MCP server to drive Pie from Claude, Cursor or VS Code without leaving the editor
- Read the Readiness Score and the deduplicated issues after each run
- Review and merge the fix pull request the Pie bot opens, or switch it to comment-only mode
Pros & Cons
Pros
- Nothing to script and nothing to maintain: no selectors, no test code, no upkeep as the interface moves
- Coverage from the first day where a scripted suite needs weeks
- Web and native mobile handled by the same tool, including Flutter, React Native and webviews
- Human review before escalation, so the queue stays short instead of full of false positives
- Fixes arrive as reviewable pull requests with root cause and visual evidence
- No SDK and no source code access, which shortens the security review
- SOC 2 Type 2 audited continuously, with a contractual commitment never to train on customer data
Cons
- No public pricing at all: quotes are custom and come only after a scan of your app
- No terms and conditions are published, so the privacy policy is the only contractual text online
- No data processing agreement is published or advertised as available on request
- Tests do not live in your repository, which creates a real dependency on the vendor
- Scope stops at end-to-end: no unit tests, no component tests, no security testing
- Only one named case study is published; the other testimonials are anonymised
- No Article 27 EU representative, and data may be processed in the United States
Pricing & Plans
There is no public pricing. PieLabs, Inc. publishes no rate card and no pricing page, and states that pricing follows the scope of your application and the coverage you want rather than a per-test unit, with a custom quote issued after a demonstration and a scan of your app. The only figure offered is comparative: less than the cost of a QA hire. A free entry point does exist, since the sign-up page advertises a free start with no credit card and setup in under five minutes, but no permanent free plan is described and no trial duration is stated. Consequently no cheapest paid price point can be quoted in any currency.
Data, GDPR & hosting
A consolidated view of how Pie handles your data.
GDPR overview
GDPR compliance is claimed explicitly and gets its own section on the security page, which cites four concrete principles: data minimisation, purpose limitation, storage limitation through ephemeral test environments, and the right to erasure on request. Section 6 of the privacy policy names the GDPR alongside the CCPA and grants EU users portability and restriction of processing. Beyond that, implementation is thin. No Article 27 EU representative is designated, no data protection officer is named, and privacy requests go to a general address. No data processing agreement is published or advertised as available. The policy also warns that data may be processed in the United States, without mentioning standard contractual clauses or any transfer framework.
Who owns the data?
No terms and conditions are published, so the only contractual document online is the privacy policy dated 24 April 2026. It states that PieLabs, Inc. does not sell personal data and shares it only with named service providers (AWS, Stripe, Google Analytics, Microsoft Clarity), with authorities where legally required, and with an acquirer in the event of a merger or asset sale. The security page adds that application data, test credentials and results are encrypted and access-controlled, that a customer's information never benefits another customer, and that test environments are destroyed after each run. Role-based access control and timestamped audit logs govern who inside your own organisation can see what.
Reuse rights
Because no terms and conditions exist, nothing on the site grants or restricts a licence to reuse what Pie produces. In practice the customer receives issue reports with screenshots, reproduction steps and visual replay links, plus fix pull requests opened against their own repository, all of which they are plainly free to use. The privacy policy grants access, correction and deletion on request and unsubscription from marketing, and the security page repeats the GDPR right to erasure. Test credentials can be added, updated or revoked at any time from the platform. Importing existing test cases by CSV is documented; a matching export path is not.
Data retention & training
Hosting summary
Hosting is only partly documented. The privacy policy names AWS as the cloud hosting provider and warns users located outside the United States that their information may be processed there, but no region, no data centre and no data residency option is declared anywhere on the site. The vendor is based in Oakland, California. What is documented in detail is the execution model: each test runs in a sandboxed environment created on demand and wiped immediately afterwards, with no persistent data and no cross-contamination between customers. Encryption is AES-256 at rest, covering application builds, test results and logs, and TLS 1.2+ in transit, with unencrypted connections refused. Test credentials sit in a separate encrypted vault opened only during active execution. Beyond the four service providers listed in the privacy policy, no formal subprocessor page and no data processing agreement are published.
Things to keep in mind
Risks and trade-offs to weigh before adopting Pie.
- Test credentials are handed to a third party; the vault is encrypted, but the scope of access still deserves an audit
- Pie tests against staging or production, so a production run can create test data or trigger real transactions
- Fix pull requests are AI-generated: merging one without serious review puts unreviewed code into production
- Tests do not live in your repository, so ending the contract erases your coverage with it
- A Readiness Score invites false comfort: 80% advertised coverage is not a guarantee that nothing is broken
- Internal testing skill can erode if the team stops writing and reasoning about its own cases
- Data may be processed in the United States with no Article 27 EU representative and no published DPA, which matters to a European data controller
Setup & Integrations
Technical difficulty
Very low for the web. There is no SDK, no instrumentation and no source code access: a URL is enough, plus credentials if the app is gated. The vendor claims setup in five minutes and a first test two minutes later, with no code to write. GitHub users get a ready-made action; any other CI needs one REST call added to an existing workflow, which most integrations complete in under five minutes. The only genuinely technical step is preparing mobile builds, and that has its own documentation for native iOS, React Native iOS and Android.
Deployment
Integrations
Supported languages
Behind Pie
Social
Resources
All the official URLs gathered for verification and reference.
Alternatives
Tools that compete with or complement Pie.
Frequently asked questions
What platforms can Pie test?
Do I have to write scripts or maintain selectors?
Does Pie need access to my source code?
Can Pie merge a fix on its own?
How many false positives should I expect?
Which tools does Pie integrate with?
How long does it take to get started?
Is my data used to train AI models?
What certifications does Pie hold?
How much does Pie cost?
Should you pick Pie?
Pie occupies a clear and unusually well-argued position: autonomous QA as a layer that runs alongside your unit and component tests rather than replacing them. The technical differentiator is genuine. Recognising interface elements visually instead of through selectors removes the maintenance tax that quietly kills most scripted suites, and adding human verification before a finding is escalated addresses the noise problem that AI testing tools usually create. Covering web and native mobile in one product is rare on this segment, and the security posture is solid for a company this young: SOC 2 Type 2 monitored continuously, ephemeral sandboxes, and a contractual commitment never to train on customer data. The weaknesses sit on the commercial and legal side rather than the technical one. Nothing about pricing is public, no terms and conditions are published anywhere on the site, and no data processing agreement is offered, which will slow any procurement team that needs to read a contract before a demonstration. The evidence base is also thin: a single named case study, and headline figures such as 80% coverage in thirty minutes that are vendor claims rather than audited results. The company is young too, with the domain registered in October 2024 and first archived that December. The natural fit is a team that ships often, has mobile in the mix, cannot justify a dedicated SDET, and is watching AI-assisted development produce more code than its current testing can absorb. Teams with a healthy Playwright suite and the engineers to maintain it will find less here, as the vendor itself says. Anyone in that first group should treat the free entry point as the honest way to judge the claims, since the product is designed to show what it finds on your own application within the first half hour.
- Choosing a selection results in a full page refresh.
- Opens in a new window.