This web agent demo is honest about being glue code

What it does Open Operator is a Next.js template that turns natural-language instructions into headless browser actions. It uses OpenAI to parse intent, Stagehand to translate that intent into DOM operations, and Browserbase to execute them in a remote browser. The entire project is essentially a thin orchestration loop sitting on top of three external services.

The interesting bit The README is unusually candid: Browserbase labels this a proof of concept, has already moved the code to a demos repository, and explicitly says they want to sell infrastructure rather than compete with agent products. The included architecture diagrams even contrast messy agent complexity with Stagehand’s tidier abstraction, which is rare honesty in a hype-heavy space.

Key highlights

• Delegates all heavy lifting: natural language to OpenAI, browser automation to Stagehand, execution to Browserbase
• Built as a Next.js application with a minimal agent loop
• Explicitly unmaintained: the source now lives in browserbase/gtm-demos
• Inspired by OpenAI’s Operator feature, but positioned as tooling rather than competition
• Open source under MIT license

Caveats

• No longer actively maintained; the repository is effectively archived
• Requires three external API keys and services to function at all
• The README warns it is only a proof of concept, not a production foundation

Verdict Good for developers who want a readable, minimal example of how to wire LLM intent into remote browser automation. Skip it if you are looking for an actively maintained framework or an agent that runs locally without cloud dependencies.