Cua Connects OpenAI Agents API to Cloud Desktop Sandboxes
- Source
- Cua
- Date
1/ Give your OpenAI agents Linux and Windows desktops. You can now connect OpenAI Agents API to Cua Cloud Fleet - with shell access, files, and Cua Driver for native desktop actions. Start a Fleet: https://t.co/39njnVAqD7
2/ OpenAI runs the agent loop. The executor runs tools inside your Cua Fleet VM. Cua Driver connects through local MCP. The agent can inspect windows and act on fresh snapshot targets. Tool results and screenshots return to OpenAI.
3/ Reconnect without starting a new workspace. The example stops the executor, sends a second input, and reconnects the original environment in the same active Fleet claim. The agent reads the file it created before the disconnect.
4/ Try the Linux guide with Cua Fleet credentials, an OpenAI API key, and a separate environment key. It covers connection, desktop actions, reconnect, file retrieval, and cleanup. Guide: https://t.co/GrcP5x2T3a Start: https://t.co/UwJdrwORDM
- OpenAI runs the while an executor runs tools inside the Cua Fleet VM. Cua Driver connects through local so the can inspect windows and act on fresh snapshot targets, and tool results and screenshots return to OpenAI.
- Cua's example stops the executor, sends a second input and reconnects to the original environment within the same active Fleet claim; the agent then reads a file it created before the disconnect.
- The Linux guide needs Cua Fleet credentials, an OpenAI API key and a separate environment key, and covers connection, desktop actions, reconnect, file retrieval and cleanup.
- agentic loop — The cycle an agent runs in: decide, call a tool, read the result, decide again — repeating until the goal is met or a stop condition fires.
- MCP — The Model Context Protocol — an open standard that lets any AI assistant plug into any tool or data source without custom integration code.
- sandbox — An isolated environment where AI-generated code or agent actions run without being able to touch anything real.
- AI agent — An AI system that doesn't just answer once but works toward a goal in a loop — taking actions, reading the results, and deciding what to do next.
OpenAI agents can now drive full Linux/Windows desktops through Cua's Fleet sandboxes with shell and file access, and sessions can be paused and reconnected without losing state.
Checking sign-in…
Loading comments…