Apogee

AI browser assistant for articles, videos, emails and more. Runs in-browser via WebGPU, WebAssembly, and Ollama.

Apogee is a private, in-browser AI assistant for articles, YouTube videos, PDFs, emails, and more. It summarizes what you're reading and answers questions about it, and it does all of this on your own machine. No cloud, no API keys, no account, no telemetry. Your page content never leaves your browser. Most "AI summarizer" extensions send the page you're reading to a company's servers. Apogee doesn't. It runs quantized language models directly inside your browser, on your GPU via WebGPU, or on your CPU via WebAssembly, so the text you summarize is processed locally and discarded. There's nothing to log in to and nothing to leak. Apogee was inspired by Mozilla's discontinued Orbit project, which offered browser summarization but routed everything through centralized cloud servers. Apogee fixes that by being fully local-first: no server dependency means no data leaks, no subscription, and nothing that can be shut down or sunset. WHAT IT DOES • Summarize any page — articles, blog posts, documentation, long threads • Summarize YouTube videos from their transcript • Summarize PDFs opened in the browser • Ask questions about the page — Apogee reads the whole page (not just the first few thousand characters) using on-device retrieval, so answers can come from deep inside a long article, PDF, or transcript • Highlight-in-page — click any bullet in a summary and Apogee scrolls to and highlights the passage of the original page it came from, so you can spot-check a claim without re-reading everything • Multiple summary formats — bullets, sentences, or paragraphs Fast ways to summarize without opening the popup: • Right-click a page → "Summarize this page" • Keyboard shortcut (default Alt+Shift+U, remappable at chrome://extensions/shortcuts) A system notification tells you when the summary is ready. TWO WAYS TO RUN IT 1. In-Browser AI (zero setup) Runs small, fast models entirely in your browser. On first use it downloads the model weights (roughly 270 MB to 2.2 GB depending on the model) and caches them locally — after that, everything works offline. Defaults to WebGPU (WebLLM) on Chrome and Edge, with a WebAssembly (Transformers.js) option in Settings for machines without WebGPU. 2. Local Ollama (for power users) Prefer larger, more capable models? Point Apogee at your own local Ollama instance and it talks to it directly over 127.0.0.1 — no separate backend to install or run. Any model you've pulled shows up automatically. Still fully local; nothing leaves your machine. PRIVACY • No cloud inference — models run on your device • No API keys, no sign-in, no account • No analytics or tracking • Network access is limited to downloading model weights (from Hugging Face) on first run, and your own local Ollama at 127.0.0.1 • Page content is processed locally and never uploaded OPEN SOURCE Apogee is free and open source (MIT licensed). Source, issues, and releases: https://github.com/darshi1337/apogee