Pig
Pig is an API and Python SDK that lets AI agents operate Windows applications through natural computer actions such as clicking, typing and taking screenshots, running on self-hosted machines. The service is no longer maintained.
What is Pig?
Pig is an API and developer toolkit that gives AI agents control of a Windows desktop. Rather than integrating with software through official interfaces, a Pig agent works the way a person does: it takes a screenshot, decides what to do, then clicks and types. That aims it squarely at the large body of Windows business software that never exposed an API and cannot easily be replaced.
The product was offered on three levels. Pig Chat put the computer-use model behind a familiar chat window, so a workflow could be connected to a machine and prototyped without writing code. The Agent API exposed the same engine programmatically. Beneath both sat a low-level Python SDK, published openly as pig-python under the Apache-2.0 licence, for teams wanting to implement their own agent loops on Pig's base toolkit.
Its architecture is unusual in one respect. Machines are self-hosted by default: rather than renting cloud desktops, a customer installs Piglet, a small open-source agent written in Zig, on their own Windows machine, which then registers as a worker in the Pig system. Automation executes on hardware the customer owns. The exposed primitives are deliberately plain, with click, type and screenshot alongside coordinate-level clicking, and a pair of calls that hand a running session to a human operator and take it back, pitched at operations too sensitive to leave unattended.
The present tense is harder. Pig came out of Y Combinator's Winter 2025 batch; its domain was registered in December 2024 and the earliest web archive capture dates from the same month. Its founder announced in May 2025 that he had moved to a different product, and the service is no longer operated. The documentation subdomain has been offline since at least April 2026, its certificate expired and its deployment gone, so every call-to-action on the homepage now leads nowhere. The API subdomain fails to negotiate a secure connection. The public repositories have not moved since March 2025. What survives is a three-page marketing site and the openly licensed code.
What it does
- Launch and operate Windows applications through an API
- Drive a desktop with native actions: click, type and screenshot
- Prototype an automation in a chat interface without writing code
- Build custom agents with a low-level Python SDK
- Run automations on your own self-hosted Windows machines
- Hand control to a human operator mid-run, then take it back
- Process batches by looping an agent task over a queue of inputs
When to use Pig / When not to
A quick filter to help you decide if Pig is the right fit.
When to use Pig
- AI engineering teams building their own computer-use agents on top of a low-level toolkit
- Developers automating legacy Windows desktop software that exposes no API of its own
- Operations teams running repetitive, queue-driven workflows across desktop applications
- Organisations that require automation to execute on machines they own and control
- Teams needing a human operator to take over an agent mid-run on sensitive steps
When not to use Pig
- Anyone needing a supported, production-ready service: the product is no longer operated
- Users on macOS, Linux, mobile or browser-only environments, since Windows is the sole target
- Buyers who need published rates before committing, as no pricing is disclosed anywhere
- Teams with compliance obligations: no privacy policy, terms, DPA or GDPR statement exists
- Non-technical users unable to self-host a Windows machine or work from code repositories alone
How to use Pig
A typical end-to-end flow, from setup to results.
- Create an account on the sign-in page, using GitHub or an email address and password
- For the no-code route, open Pig Chat, connect a machine and describe the workflow in the chat
- For the developer route, download Piglet and run it on a Windows machine you control
- Confirm Piglet is listening locally, on the port it exposes on that machine
- Retrieve an API key from the account area of the site
- Install the Python SDK with pip, adding the pig-python package to your project
- Import the client and select the target machine by its identifier
- Open a session on the machine and drive it with screenshot, click and type calls
- Alternatively, delegate to the agent and give it a task in plain language
- Where oversight is needed, yield control to a human operator, then await its return
Pros & Cons
Pros
- Addresses a genuinely under-served problem: desktop Windows software with no API
- Self-hosted by default, so machines and data stay on hardware the customer controls
- SDK and worker agent are open source under permissive licences, so they outlive the service
- Three levels of entry, from a no-code chat interface to a low-level toolkit
- Human handover built in, which matters for operations that should not run unsupervised
- Primitives are simple and readable, modelled directly on ordinary human gestures
Cons
- The product is abandoned: its founder moved to a different venture in May 2025
- Documentation is offline, so every call-to-action on the homepage leads to a dead end
- The API subdomain no longer completes a secure connection, breaking any live integration
- No pricing, no plans and no free trial are published anywhere
- No privacy policy, terms of service, legal notice or GDPR statement exists
- Windows only, with no support for macOS, Linux, mobile or browser environments
- The SDK was self-declared alpha, with breaking changes and no repository activity since March 2025
Pricing & Plans
No pricing is published. Pig operates no pricing page, and no amount, plan or currency appears anywhere on the site. Neither a permanent free plan nor a free trial is announced. Self-service registration exists through GitHub or an email address, but without any published rate card; commercial arrangements, including implementation services and priority support, were handled individually through the contact form.
Data, GDPR & hosting
A consolidated view of how Pig handles your data.
GDPR overview
There is no mention of the GDPR anywhere on the site. Pig publishes no privacy policy, no terms and conditions and no legal notice, and the usual legal routes return errors. No data controller is identified, no lawful basis is stated, no procedure for exercising data subject rights is described, no Article 27 EU representative is designated and no data protection officer is named. No email address of any kind is published; the only channel is a web form. This is a complete absence of documentation rather than a partial or informal compliance posture. Nothing on the site allows the level of GDPR implementation to be assessed, and nothing indicates that any was undertaken.
Who owns the data?
No answer can be drawn from the vendor's own documents. Pig publishes no terms and conditions, no privacy policy and no legal notice, and the site consists of only three pages. Nothing therefore states who owns the material an agent handles, whether screenshots, keystrokes or the contents of the Windows applications it drives, nor what the vendor may do with it, nor which third parties might receive it. The one mitigating fact published is architectural: worker machines are self-hosted by default, so execution takes place on hardware the customer controls. Ownership itself remains entirely undocumented, and no contractual commitment on the point can be cited.
Reuse rights
The site publishes no terms and conditions, so nothing states whether the end user may reuse what Pig produces, under what licence, or subject to what restrictions. No permission regime, no attribution requirement and no clause on commercial use is documented anywhere on the site. What can be established is factual rather than contractual: automations run on machines the customer self-hosts and registers as workers, so outputs are generated on the customer's own hardware. That is a consequence of the architecture, not a right granted by the vendor. Anyone needing certainty about reuse would have had to obtain it directly through the contact form, the only channel the site offers.
Data retention & training
Hosting summary
No hosting jurisdiction is announced for customer data. The site names no country and no region, and publishes no privacy policy or security page where such a commitment would normally appear. Three distinct layers should not be confused. The marketing site itself is served from a content delivery network on an address geolocated in the United States, which concerns the web pages and not the automation data. The API subdomain pointed at a third-party compute platform, though it no longer responds. The working machines are self-hosted by the customer, so their location, and therefore the jurisdiction of anything processed on them, is determined entirely by the customer rather than by the vendor. Nothing published allows any further statement about where data was held, under which law, or with which safeguards for international transfers.
Things to keep in mind
Risks and trade-offs to weigh before adopting Pig.
- The service is no longer operated: do not build or migrate a project onto it
- A computer-use agent sees the whole screen and acts as the logged-in Windows user, which is a wide opening on business applications
- The API subdomain no longer responds, so any surviving integration is very likely broken
- No contract of any kind is published: no terms, no privacy policy, no DPA, no GDPR statement
- No email address exists, so an incident leaves only a web form as recourse
- The toolkit was self-declared alpha, with breaking changes announced without warning
- The domain is registered only until December 2026, after which the site itself may disappear
Setup & Integrations
Technical difficulty
Two very different levels. The no-code route is light: create an account, connect a machine and prototype the workflow in the chat interface. The developer route is substantially harder: you need a dedicated Windows machine, the Piglet worker installed and listening on it, an API key, the Python package installed, and an agent loop written around the client. Today both routes are complicated by the documentation being offline, leaving only the public code repositories as guidance.
Deployment
Supported languages
Behind Pig
Resources
All the official URLs gathered for verification and reference.
Frequently asked questions
What does Pig do?
Which platforms does it support?
Do I need to write code to use it?
Where do the automations actually run?
What can the agent do on the desktop?
How much does Pig cost?
Is there a free plan or a free trial?
Is the code open source?
Can I still read the documentation?
Is Pig still being maintained?
Should you pick Pig?
Pig set out to solve a genuinely awkward problem. A great deal of business software runs only on Windows, was written before anyone thought about APIs, and is too entrenched to replace. Handing an AI agent a keyboard and a screen is a reasonable answer, and Pig's version of it was carefully built: three levels of entry from chat to low-level toolkit, self-hosted workers so the machines stay with the customer, and a human-handover mechanism for steps that should not run unsupervised.
It did not find its market. The founder moved to a different product in May 2025, and the consequences are now visible from outside: the documentation subdomain has been dark since at least April 2026, the API subdomain no longer completes a secure connection, and the public repositories have not been touched since March 2025. Every button on the homepage points at documentation that is gone.
Anyone evaluating Pig today should treat it as a closed service rather than a product to adopt. There is no pricing to compare, no terms of service, no privacy policy and no support channel beyond a contact form. The domain itself is registered only until December 2026, so even the marketing site may not persist.
What remains still has value. The Python SDK and the Piglet worker agent are published under permissive licences and can be read, forked and reused, which is more than most discontinued services leave behind. For a team building Windows computer-use tooling of their own, Pig is best regarded as prior art worth studying: a well-reasoned design whose infrastructure is gone but whose code is not.
- Choosing a selection results in a full page refresh.
- Opens in a new window.