v0.1.2468 Sep 2026·Privacy & Shields, Memory & reading, Tabs & interface

Memory: your browser remembers what you read, and can prove it

This is the release Floati was built for. Your browser now remembers the pages you actually read, on this device and never in a cloud, and lets you ask questions of that memory in plain language: you get back the exact sentences you read, with every source cited so one click takes you back to it. Starting from empty is optional. You can bring in the history you already have in another browser, addresses and titles only and all on-device, then let Floati quietly re-read those pages in the background so it can answer from what they said, not just their names. And because Floati paints pages with its own engine, where the same page always produces the same pixels (something a browser that draws on the graphics card cannot promise), it can seal a page into a small file anyone can reopen and check byte for byte. Every answer you get can be reopened exactly as it was served, tamper-evident, and any page can be sealed on demand to keep or to share as proof. Recall and proof now live together on one page. None of this leaves your machine.

Parity with the reference browser

74overall, as of v0.1.343

Web APIs / DOM surface42

Security architecture66

GPU / graphics acceleration68

DevTools / extensions72

Networking85

Media85

Accessibility85

Real-world site compatibility99

HTML parsing100

CSS parsing + selectors100

Layout: block / flex / grid / tables / writing-modes100

Paint + visual100

JavaScript language100

JavaScript performance (JIT)100

HTML parsing100

What changed 6 changes

Memory & reading	Ask your reading, from the full text. Type a question in plain language and Floati answers from the pages you have actually read, quoting the most relevant lines and listing every source so a click takes you straight back. It searches only your own on-device reading archive: no server, no outside model, nothing sent anywhere. As you browse, the pages you read are indexed locally, so the answers get richer over time.
Memory & reading	Seed it from the browser you already use. So the memory is not empty on day one, you can import the visit history from the browser you used before: addresses and titles only, read once and kept on your device. An optional background pass then re-reads those pages one at a time while you are idle, and only while you are idle, so Ask can answer from their full text; it is bounded, stoppable, and honest that a page fetched today can differ from what you saw back then.
Memory & reading	Verifiable recall: reopen any answer exactly as it was served. Every cited source can be reopened as a Web Capsule, a sealed offline snapshot of the page as it was when you read it, with an honest badge saying whether the original bytes are intact or have been altered. This is possible because Floati's engine is deterministic: the same saved bytes always repaint the same pixels, so what you saw can be shown again and checked, which a browser whose output depends on the graphics driver cannot do.
Privacy & Shields	Seal any page as proof, and verify one offline. On demand, capture a live page into a small .flcap file carrying a Render Receipt: a single fingerprint that binds the address, the exact bytes the site served, and the exact pixels Floati drew from them. Publish the fingerprint, hand someone the file, and their own copy of Floati re-derives the same fingerprint if nothing was altered, fully offline. Honest by design: it proves what was rendered from the saved bytes, and the tools spell out what it does and does not attest.
Tabs & interface	One page for memory and proof. The passive memory (Ask your reading) and the deliberate, file-based sealing and verifying now share a single page, laid out like the password vault: a question-and-answer hero, a place to seed your memory, and a seal-or-verify tool, with a side rail that explains plainly how recall works, why it is trustworthy, and where its limits are.
Memory & reading	Rendered by Floati's own engine from the first paint on macOS. Normal browsing on macOS now draws with the from-scratch engine from the very first frame, with the system renderer kept as an automatic per-site fallback for anything the engine cannot yet draw. This is what lets memory and receipts happen quietly in the background as you read, and capsules are captured at full fidelity from real page renders rather than a separate snapshot.