Apogee is a privacy-first, offline-capable browser extension that performs AI summarization for web pages, videos, PDFs, DOCX, and pasted text entirely on the user’s device. It runs models using WebGPU on compatible Chromium browsers or WebAssembly as a universal fallback, and can also connect to local model servers (Ollama or llama.cpp) over loopback for larger models. Designed to replace a server-dependent predecessor, Apogee intentionally avoids cloud APIs, accounts, telemetry, or remote storage: models download once and are cached, tokenization, inference, retrieval, and caching all occur locally, and local loopback connections are strictly limited to 127.0.0.1.
Feature-wise, Apogee extracts clean article text (with site-specific extractors), summarizes selected text, builds interactive timestamped timelines for YouTube/Bilibili, parses social threads into structured Markdown, and supports grounded sentence highlighting that jumps to source passages. It offers persistent side panels, custom standing instructions, per-page focus keywords, multi-language translation via Opus-MT, and sensitive-site exclusion for disk caching. The project is open source (MIT), distributed as Chrome/Firefox extensions, and includes extensive documentation and developer tooling: architecture, model references, privacy/security details, and an extractor test harness for contributor workflows.
Summary generated by AI from the linked article. hn.today is not affiliated with Hacker News or Y Combinator.
AI agents can now draw large arrows, boxes, and text on Mac screens using a simple CLI tool. The feature is compatible with Claude Code and Codex, and is released under the MIT license. (github.com)
OpenAI fired three safety researchers for allegedly mishandling sensitive information, but they dispute the claims and warn it creates a chilling effect on internal communication. The researchers argue that their dismissal signals a shift away from an open safety culture at the company. (techcrunch.com)
Hugging Face tests a memory layer called galahad-kv that stores the key-value state of text blocks on local NVMe disks, enabling large language models to handle up to 50 million tokens. This approach avoids recomputing internal states, improving efficiency for very long prompts. (huggingface.co)
OpenTelemetry introduces native design principles that enable products to export telemetry data to any observability stack seamlessly. The update emphasizes compatibility, scalability, and new features across various SDKs and agents to improve observability infrastructure. (opentelemetry.io)
Users discuss the advantages of subscribing to hosted LLM services versus deploying open-source models locally. They share reasons related to cost, convenience, and customization options. (news.ycombinator.com)
Four coding agents were given a $100 budget to build a functional PDF editor, but their implementations had noticeable bugs and shortcomings. The experiment highlights that current AI agents struggle with goal pursuit and user interaction, leading to subpar software quality. (blog.nielstron.de)
Today's best Hacker News stories, summarized and screenshotted, one email a day.