Unwall

The problem

Reading tools solve the layout problem and create a privacy one. The usual shape is to ship the page you are reading to a server, render it there, and send back something cleaner. That works, and it means a third party now holds a list of everything you read closely enough to want help with. Unwall takes the harder path: everything happens in the tab.

PDFs were the decision that set the cost

Cleaning up an HTML page in the browser is ordinary work. PDFs are not — they are a layout format, not a text one, and pulling readable prose back out of one takes a real parsing engine.

The cheap answer is to upload the file and let a server do it. Unwall bundles the PDF engine instead and runs it locally, which costs bundle size on every install so that no document ever leaves the machine. Password-protected files follow the same rule: the password opens the document in the tab and is never written to preferences, never kept with the recovery data, and never sent to the studio.

The one place the network is unavoidable is a PDF referenced by address rather than picked from disk. Unwall asks for permission for that specific origin and fetches it, which means the host of that document sees a request. That is stated plainly in the privacy policy rather than glossed over.

Read-aloud, where accessibility and privacy pull apart

Unwall speaks through the browser’s own speech synthesis and highlights each word as it goes. Depending on the voice, that speech is generated either on the device or by an online provider — and the online ones usually sound better, so automatic selection can reach for one.

An online voice receives the text being spoken. That is a genuine conflict: the most accessible option is the least private one. Unwall resolves it by naming the trade-off and leaving the choice with the reader, who can pin a device voice and keep the text local. Quietly defaulting to the nicer-sounding voice and staying silent about where the words go would have been easier and worse.

Restoring a document without becoming a history

Refreshing a tab should not cost you a fifty-page PDF. Unwall keeps the extracted text, the title, and page previews in local IndexedDB so a document can come back, with a tab-scoped reference in session storage to tie it to the right place.

Snapshots stay restorable for twenty-four hours. Cleanup runs when an expired record is touched or when a new document is saved — it is not a background timer, so an expired record nobody goes near can sit on disk longer than a day. The privacy policy says exactly that, because a retention promise you have not actually implemented is just a nicer lie.

Webpage reading never touches disk at all; it uses an in-memory cache that dies with the tab. Unwall reads the current tab’s address to tell whether a page is supported, and does not assemble that into a record of where you have been.

The decisions

Parsing

PDF engine bundled and run locally, not server-side

Passwords

Used in the tab, never persisted or transmitted

Speech

Browser synthesis; online voices disclosed, device voices selectable

Retention

PDF snapshots restorable 24h; cleanup behaviour documented honestly

Collected

No account, no advertising, no usage analytics