Demo video
Set videoUrl on this product in config/shipped.ts
Overview
You learn across a dozen places — YouTube, blog posts, ChatGPT threads, PDFs — and almost none of it survives the week. Atlas is a memory layer for that: one click captures whatever you're looking at, and the system organises and recalls it later. It's deliberately not another note-taking app; nothing has to be filed by hand.
The architecture is built around one constraint: capture is synchronous and fast, intelligence is asynchronous. The capture endpoint returns inside 300ms and never calls a model on the request path — summarising, embedding and extraction all happen afterwards in Redis-backed workers. A Chrome MV3 extension does the capturing, a FastAPI service handles persistence, and a Next.js app is the dashboard you read it all back in.
Working today
- One-click capture of pages, videos, AI conversations and PDFs
- Chrome MV3 extension built with WXT and React
- Capture returns in under 300ms, with no model call on the request path
- Auth, onboarding and the web dashboard shell
- Type-safe API client generated from the OpenAPI spec
- Playwright end-to-end tests covering the capture flow
Planned
- AI summaries of everything captured
- Knowledge that self-merges across sources
- Natural-language search with citations back to the source
- Coverage tracking to surface gaps in what you've learned
- Spaced repetition and quizzes
- In-browser suggestions while you read