From model to agent: Equipping the Responses API with a computer environment

Responses API was equipped with a computer environment, openAI added a hosted container workspace and shell tool to the Responses API.

Shell tool executes shell commands proposed by models the model proposes commands and the shell tool runs them in an isolated environment.

Container runtime receives shell commands from the Responses API and returns streamed output the API forwards commands to the container runtime and relays output back to the model in near real time.

Models GPT‑5.2 and later are trained to propose shell commands the prompt must mention using the shell tool and the selected model must be trained to propose shell commands.

Responses API supports concurrent execution of multiple shell commands the API can run commands concurrently using separate container sessions and multiplex streams back to the model.