Hey HN. I built Unbrowse because browser automation for OpenClaw agents is absurdly slow.
Eg: when your agent visits Polymarket, it launches Chrome, waits 12 seconds, and scrapes the DOM. But the page itself called GET /api/markets in 200ms and got clean JSON back.
The browser is just rendering JSON into HTML so the agent can scrape it back into data. That's insane.
Unbrowse captures the internal API calls via CDP, extracts endpoints + auth, and generates reusable API skills. Browse once, direct API calls forever. 200ms vs 12 seconds.
Skills are shared - one agent learns a site, they can optionally monetize it when publishing, and other lobsters can benefit.
Basically browser-use skills, but open-source and integrated into your virtual lobster.
Hey HN. I built Unbrowse because browser automation for OpenClaw agents is absurdly slow.
Eg: when your agent visits Polymarket, it launches Chrome, waits 12 seconds, and scrapes the DOM. But the page itself called GET /api/markets in 200ms and got clean JSON back.
The browser is just rendering JSON into HTML so the agent can scrape it back into data. That's insane.
Unbrowse captures the internal API calls via CDP, extracts endpoints + auth, and generates reusable API skills. Browse once, direct API calls forever. 200ms vs 12 seconds.
Skills are shared - one agent learns a site, they can optionally monetize it when publishing, and other lobsters can benefit.
Basically browser-use skills, but open-source and integrated into your virtual lobster.