deta Surf

Research work tends to scatter across browser tabs, downloaded PDFs, video timestamps, and a separate notes app, with constant copy-pasting in between. deta Surf, from the team behind Deta, is a desktop application that tries to collapse that workflow into one place. It calls itself an AI notebook: you pull local files and web content, including web pages, YouTube videos, tweets, and PDFs, into notebooks that live on your machine. The central idea is local-first. Your library is stored on your own computer in an open, transparent format rather than on a vendor's servers, and you can read most of it offline. On top of that sit AI features for note generation, search-linked citations, and small generated apps the project calls Surflets.

The app is built with Svelte, TypeScript, and Rust, and runs on macOS, Windows, and Linux. Data is kept in what the project calls SFFS, the Surf Flat File System, an open on-disk format meant to keep your content portable. You bring almost any media into a private library, then organize it into notebooks. Smart Notes can draft content from a website, file, or library item via @-mentions, and the citations are deeplinked back to the source: a specific PDF page, a video timestamp, or a web page. What stands out most is model flexibility. You can supply your own API keys for popular cloud models or run local models through Ollama, so the AI layer is not tied to a single provider. Surflets generate small interactive apps to explore data without writing code.

deta Surf is aimed at students, researchers, and knowledge workers who read and synthesize a lot of source material and want it kept on their own machine. Concrete uses include taking structured notes on long YouTube lectures with clickable timestamps, working through multi-hundred-page PDFs and jumping back to the exact cited page, gathering web research into a single notebook instead of a wall of tabs, and parsing scientific diagrams or foreign-language documents. The local-first design appeals to people handling sensitive or private material that they would rather not send to a cloud notebook by default. The bring-your-own-key and local-model options suit cost-conscious users and tinkerers who want to choose between a frontier model and a small offline one depending on the task.

The honest caveats start with maturity. deta Surf is an open beta, currently around version 1.4.7-beta, so behavior, file formats, and features can still shift, and it is not the choice for production-critical workflows. No pricing has been communicated, which leaves the long-term monetization model unclear: the code is open under Apache-2.0, but whether hosted services or paid tiers arrive later is unknown, and that matters for anyone planning years ahead. Some marketing language, such as references to indexing huge volumes of pages, is not measurable and should be read as aspiration rather than benchmark. Bring-your-own-key means the AI features still cost money through whichever provider you use unless you run everything locally, and local models trade quality for privacy. As with any young desktop app, expect rough edges and gaps in documentation.

Against the obvious commercial reference points, such as NotebookLM, Notion AI, or browser-integrated assistants, deta Surf's distinguishing bet is ownership: your data stays local, the format is open, and you choose the model. That is a genuine difference for privacy-sensitive research, and the Apache-2.0 license plus an open-core repository make it inspectable and forkable in a way hosted products are not. It is worth adopting now if you do research and note-taking with local data, want free choice of models, and are comfortable living on a beta. It is too early if you need a stable, supported tool for mission-critical work, or if predictable long-term cost and a clear commercial roadmap are decisive. For curious researchers, it is a credible, low-lock-in option to try today.