Introducing AI Extract (Beta)

FlowFerry has always extracted articles the classical way: a rule-based reader (we use Defuddle, the engine behind Obsidian's clipper) looks at the page structure and pulls out what looks like the article. It is fast, it runs entirely on your device, and for article-shaped pages it works very well.

But the web is less and less article-shaped. A xiaohongshu note keeps its photos in a carousel widget far away from its caption. An X thread is a pile of nested divs. Lazy images, infinite comment sections, pages that are really apps: structure-based rules can only guess, and sometimes they guess wrong.

AI Extract is our answer, as an optional beta. When it is on, FlowFerry lets a language model read the page the way a person would, and keep what matters.

How it works

1. The app loads the page in a hidden browser, scrolls it, and waits for lazy images to reveal themselves, so the model sees what you see.
2. If we know the site, we cut the page down to its main content container first. Less noise in, better article out.
3. The trimmed page goes to our API, which calls a model through Vercel AI Gateway. The extracted article streams back as it is generated, so you watch it appear in the preview instead of staring at a spinner.
4. If anything goes wrong (no connection, quota reached, the model has a bad day), FlowFerry silently falls back to the local extractor. Saving never breaks.

The result is a normal FlowFerry article: stored on your device, readable offline, exportable, sendable to Notion or Obsidian like everything else.

What about privacy?

This is the part we care most about, because "local-first" is the whole point of FlowFerry.

AI processing runs with zero data retention. The page content is processed in memory, is never stored on our servers, and is not logged or used for training by us, by Vercel, or by the model provider. The only thing we record is that your account used one extraction at a certain time, so we can enforce the usage limits below. We do not know what you saved, and we like it that way. The details are in our privacy policy.

Limits and pricing

Every extraction costs us real inference money, so the beta ships with honest limits:

• Free accounts get 10 one-time trial extractions.
• Pro accounts get 20 extractions per day.

Everything else about FlowFerry's pricing is unchanged.

How to try it

Update to the latest version, open the Home tab, tap the plus button next to the input field, and switch on AI Extract (Beta). You need to be signed in. Then save something hostile, a photo-heavy social post or a page your usual clipper mangles, and watch it come out readable.

It is a beta: the output is machine-generated and occasionally imperfect, and simple pages are still faster through the local extractor. Tell us what breaks at hi@flowferry.app, and happy reading.