Apogee
Private, in-browser AI summarizer for articles, videos, and PDFs that runs entirely on-device via WebGPU, WebAssembly, or a local Ollama instance.
Open source alternative to:
Apogee is a privacy-first browser extension that summarizes articles, YouTube and Bilibili videos, and PDFs directly inside the browser. All inference runs on-device — on the GPU via WebGPU in Chromium browsers, on the CPU via WebAssembly (the Firefox default and an opt-in fallback on Chromium), or against a user-hosted Ollama server on the loopback interface.
Key Features
- Articles and web pages: Clean text extraction via Mozilla Readability and specialized site extractors.
- YouTube and Bilibili videos: Interactive timestamped summaries you can click to seek directly into the video.
- Client-side PDF support: PDF text is parsed locally without external conversion servers.
- Ask Q&A with local retrieval: Embedded passages are matched on-device so you can ask follow-up questions about long documents without losing context.
- Grounded sentence highlighting: Click any summary bullet to scroll the page to the original source passage (Chromium).
- Custom standing instructions: Set personal prompt guidance like explain simply or be technical.
- Multi-language translation: Summarize pages into 29 target languages via Helsinki-NLP Opus-MT or direct LLM translation.
Why it replaces cloud AI extensions
Most AI summarizer extensions (Merlin AI, HARPA AI, MaxAI) route the page text to third-party cloud APIs, require API keys or paid subscriptions, and break without an internet connection. Apogee inverts that model: weights are cached locally after the first run, so it keeps working offline and never sends page content, transcripts, or summaries off the device. Power users can opt into Ollama mode to run larger models against a local instance on http://127.0.0.1:11434, still with nothing leaving the machine.
Apogee is explicitly a spiritual successor to Mozilla discontinued Orbit project, designed without Orbit reliance on remote Mistral 7B servers and server-side summary caching.
Categories:
Tags:
Frequently Asked Questions
Which browsers does Apogee support?
Apogee ships for Firefox via Mozilla Add-ons and for Chrome, Edge, and other Chromium browsers via the Chrome Web Store. Chromium builds default to WebGPU; Firefox defaults to WebAssembly. WebAssembly is also available as an opt-in fallback on Chromium machines without WebGPU support.
Does Apogee require an API key or account?
No. Apogee downloads and caches model weights locally after the first run, then performs inference on-device. No account, no API key, no subscription is required for the default WebGPU or WebAssembly modes.
How does Apogee handle YouTube and Bilibili video summaries?
Apogee pulls the video transcript locally, segments it, and produces an interactive timestamped timeline. Clicking a bullet point seeks the video player directly to that point in the video on supported sites.
Can I use a larger local model with Apogee?
Yes. Apogee can connect directly to a local Ollama instance over the loopback interface on http://127.0.0.1:11434. This lets power users run larger models such as Llama or Mistral variants on their own hardware, with page content still never leaving the machine.
Does Apogee work offline?
Yes. After the first run downloads and caches the model weights, Apogee continues to summarize articles, videos, and PDFs without an active internet connection. The optional SponsorBlock lookup on YouTube uses k-anonymity and can be disabled under Settings to stay fully offline.
Looking for contributors
This project is actively seeking help, join the community!
Repository details
Updated 8/25/2026, 8:00:13 AM
View RepositorySponsor Apogee
Sponsor Apogee on GitHub Sponsors
https://github.com/sponsors/darshi1337