Open Source Radar — September 8: the browser is the new agent runtime
Today's board is unusually coherent — three of the four hottest repos are about getting an agent onto the open web without paying Chrome's memory tax or getting fingerprinted at the door. The fourth is a research agent that wants to do your reading for you.
Lightpanda (Zig, 35,096 stars) — A headless browser written from scratch in Zig rather than forked from Chromium, aimed squarely at AI agents and automation. The pitch is measurable: its own benchmark across 933 real pages puts peak memory at 123MB against roughly 2GB for headless Chrome, and five seconds to load 100 pages against 46. It speaks the Chrome DevTools Protocol, so Playwright and Puppeteer scripts mostly just work. Version 0.4.0 landed on August 31. If you're running dozens of concurrent agent sessions, the memory difference is the difference between one box and ten.
camofox-browser (JavaScript, 10,118 stars) — The opposite bet: instead of a leaner engine, a stealthier one. It wraps Camoufox, a Firefox fork that spoofs fingerprints at the C++ level — hardware concurrency, WebGL renderers, screen geometry — so the tell never reaches JavaScript, which is why most stealth plugins eventually fail. On top of that it adds the parts agents actually need: stable element refs for clicking, accessibility snapshots roughly 90% smaller than raw HTML, per-user session isolation, and search macros for common sites. It idles at around 40MB, so it can share a small box with the rest of your stack. Use it when the pages you need are behind a bot check.
feynman (TypeScript, 9,152 stars) — An open-source research agent that does the boring parts of literature work: it searches papers and the web, scores what to read first using citation, method, reproducibility, and provenance evidence, writes literature reviews with consensus and open questions, and can audit a paper's claims against its public codebase. It runs as a terminal app or a local workbench, and it works with local models through LM Studio, LiteLLM, Ollama, or vLLM if you don't want to send your reading list to a hosted API. For anyone drowning in arXiv tabs, this is the most practical research agent on the board.
archify (JavaScript, 53,899 stars) — An agent skill that turns a codebase or a plain description into a polished, interactive system map. Agents emit a typed intermediate representation and Archify compiles it deterministically into self-contained HTML and SVG, so the diagram can't drift into something the code doesn't say — every node is traceable back to source. It also diffs two validated snapshots as before, delta, and after, with exact added, removed, and rerouted facts. That makes it usable for reviewing architecture changes before merge, not just for making a pretty README graphic.
Worth watching this week: agents are spending less time in the terminal and more time on the live web, and the bottleneck has quietly moved from model quality to how cheaply and invisibly your agent can open a page.
Would you rather ship a leaner browser or a stealthier one — and what's your agent actually scraping? Tell us in the comments.
Sources: Lightpanda (GitHub) · Lightpanda benchmarks · camofox-browser (GitHub) · feynman (GitHub) · archify (GitHub)
