Deploy agents with managed deployments

A managed worker runs your Agents SDK code on infrastructure hosted by Mistral. Define and test the agent locally, then deploy its project with the SDK CLI. You do not need to operate a worker process yourself.

The maintained deployment procedure lives in Agents SDK run modes: managed worker.

• Complete SDK installation and the quickstart.
• Use an API key with workflows enabled.
• Prepare the agent project, its dependencies, a Dockerfile, and a .dockerignore that excludes credentials and local files.
• Commit and push the project to the Git repository used for the deployment.

Follow the managed-worker procedure to configure the project, run agents deploy, and inspect rollout status and logs. The guide also explains the container's agents run-worker command, agent naming, and hardening.

For credentials, follow the guide's secret bindings. Bind references to existing workspace secrets, not secret values, and keep the complete set of bindings in each deployment. Avoid baking credentials into image layers.

Worker-backed sessions and Studio tracing are covered in the same guide.