browser-use

Homonym warning. This is the GitHub org and its agent framework. anthropic's Browser Use toolset in the Claude API is claude-browser-use — a different thing with the same two words and no connection between them. The reporting on the Anthropic tool (anthropic-browser-use-tool) calls the overlap "slightly confusing" and says the same.

The GitHub org behind the browser-use agent framework (LLM agents that drive a web browser) and, under the same roof, video-use — conversational video editing as an agent skill. Both ship MIT.

The pairing is the interesting part. A browser-automation team's second product is a video editor, and the through-line is method rather than domain: give the agent a structured text representation of a medium instead of raw frames or pixels, then look at the medium only at decision points. That is the same argument mobilerun makes for phones from a completely unrelated org, which is why text-first-agent-design is paged as a thing rather than a quirk of one repo.

The browser framework itself is unsourced here — a page candidate if a source arrives, and an obvious adjacency to ../web-browsers-wiki (which owns agentic browsing as a product feature). It drives Chromium through the Chrome DevTools Protocol and brings its own agent loop and browser abstractions, which is the detail that makes it incompatible with the Anthropic toolset rather than merely unrelated to it anthropic-browser-use-tool.

A third, unrelated party in the naming space: auto-browser also drives Chromium over Playwright, but speaks MCP rather than the Chrome DevTools Protocol directly, and adds a human-takeover UI (noVNC) plus signed audit receipts that neither this framework nor claude-browser-use documents.

Related

video-use · text-first-agent-design · mobilerun · claude-browser-use · auto-browser