Skip to main content
Read the Notte skill and follow its setup instructions when setting up Notte or building browser automation. This is the shared source for CLI setup, authentication, and the browser-to-code workflow.
OpenAI Computer Use returns browser actions from screenshots. These Python and TypeScript examples connect its native Responses API computer tool to a Notte browser through Playwright over CDP.

​
Set up

Clone the Notte repository and open examples/openai-computer-use. Its README contains the complete runnable projects. Set NOTTE_API_KEY from Notte and OPENAI_API_KEY from OpenAI. Real model requests incur OpenAI API charges.
Use Python 3.11+ and uv. Run uv sync --locked, then uv run python python/demo.py.

​
Run a visual browser task

The example uses gpt-6.1-sol to read a random code from a screenshot and submit it through a form. It verifies the actual page result and returns output, session_id, and the actions executed. Each run owns its Notte session and closes it on success or error. The reusable adapter executes ordered action batches and returns a screenshot at the original viewport resolution. It supports mouse movement, clicks, scrolling, dragging, typing, key presses, waits, and screenshots within the browser. Provide a permission callback when adapting the adapter to other pages. The demo approves actions only on its own test page; pending API safety checks stop execution for human review. Keep consequential actions under your user’s control, and treat page content as untrusted. Runs have turn, action, and time limits.