Why Unwrite isn't open source

People ask this. It is a fair question. If we bang on about privacy and "your content never leaves your browser," why not open source the whole thing so people can verify it?

Here is the honest answer.

Open source does not equal private

Open source means the code is available to read. It does not mean the code respects your privacy. Plenty of open source projects include telemetry, analytics, and tracking. The code is right there in the repository for anyone to read, and most people never do.

The reverse is also true. Closed source software can be genuinely private. What matters is architecture, not access to a Git repository.

Your browser is the audit tool

You do not need our source code to verify our privacy claims. You need a browser.

Open DevTools. Click the Network tab. Use any Unwrite tool. Watch the requests.

Be precise about what you're looking for, because "no requests at all" was never the claim and would be a strange thing to promise. You will see the page fetch its own code: JavaScript and WASM modules from our origin and from jsDelivr, and model weights from Hugging Face if you're using the in-browser LLM. Those are inbound. What you will not see is a request carrying the file or the text you just gave the tool. That is the claim, and the Network tab settles it in seconds.

It's a stronger guarantee than reading source code, because source code can differ from what is actually deployed.

The network tab shows you what the live site actually does. Not what a repository says it should do. Not what a commit from six months ago intended. What it does right now, on your device, in your browser.

The "community reviews it" myth

The strongest argument for open source security is peer review. The community reads the code, finds bugs, and submits fixes. In theory this works brilliantly.

In practice it fails at scale. Consider two of the most significant security failures in recent open source history:

Heartbleed (2014): A critical vulnerability in OpenSSL, one of the most widely used open source libraries on the internet. The bug existed in the public codebase for two years before anyone noticed. Millions of servers were exposed. The code was right there. Nobody looked.

Log4Shell (2021): A remote code execution vulnerability in Log4j, a Java logging library used by nearly every enterprise Java application. The vulnerable code was public for years. The fix came from a security researcher, not from routine community review. The exploit was trivial. The exposure was catastrophic.

Open source does not guarantee that anyone competent is reading the code. For most projects, the number of people who have actually audited the security-critical paths is effectively zero.

Handing over the playbook

This is the part people tend not to think about. Open sourcing a tool like Unwrite means publishing every detection heuristic, every transformation rule, and every signal weight that our AI detection and humanisation system uses.

That is a roadmap for anyone who wants to defeat it.

We are not being hostile about this. We think AI tools are useful and people should use them. But the detection and humanisation system has value precisely because the internals are not public knowledge. Making them public does not make the tool better for users. It makes the tool worse.

Sustainability

We are a small team. The tools are free. There is no subscription paywall on any of the browser-based tools and there will not be one.

Open sourcing changes the economics. It creates an expectation of community management, issue triage, pull request review, and documentation maintenance. Those are real costs that do not generate revenue. For a project that already gives away its core product for free, adding open source maintenance overhead would accelerate burnout without improving the product for users.

Where we are open source

This is not a blanket objection to open source. We contribute to open source and have published open source work.

Unwrite Images is built on Squoosh, Google's open source image compression toolkit. Our version extends it with batch processing, additional format support, and integration with our tooling. The work is published at github.com/benpalmer1/Unwrite-Images.

Where open sourcing makes the tool better for users and does not undermine its purpose, we do it. The image tools benefit from community scrutiny because the processing logic is well-understood and there is no adversarial concern. Anyone can verify the compression algorithms produce correct output.

What we do instead

Instead of publishing source code and hoping someone reads it, we design the system so that privacy is architectural.

• All processing runs in your browser via WebAssembly and JavaScript
• No file uploads. Your documents are never sent anywhere for processing
• No analytics beacons, no tracking pixels, no cookies
• No user accounts required for any of the browser tools on unwrite.co. (Unwrite Voice is a separate paid product on its own subdomain, it generates on a server, and it does have sign-in. Different product, different deal, stated plainly.)
• DevTools Network tab verifies all of this in seconds

This is what we call "architectural privacy." The system cannot violate your privacy because it is not built with the capability to do so. There is no endpoint your documents can be sent to. There is no analytics SDK to fire. There is no tracking cookie to set.

Four narrow edge endpoints do exist, and we would rather name them than let you find them in DevTools: a fetch proxy and an image proxy that let the social share preview tool read public pages your browser is not allowed to read directly, the optional sign-on-your-phone relay, and an opt-in proxy for remote LLM providers that block browser requests. None of them touch the files you process. The security page sets out exactly what each one sees.

Read our privacy policy for the full details, and our post on removing all tracking from the site for the lengths we go to.

The real question

The question is not "is the source code public?" The question is "can I trust this tool with my data?"

You can answer that one in under a minute. Open DevTools. Use the tool. Watch the Network tab. Set aside the requests fetching code and models, from our origin and jsDelivr, because every site on the internet fetches those. Then look for a request carrying what you just put in. There isn't one. That is a stronger privacy guarantee than any open source licence provides.

We think that is more honest than publishing a repository and implying that makes you safe.