Build and deploy an agent end to end

This journey chains the Studio quickstarts along the loop: build, add context, evaluate, deploy, observe, improve. Each step lives in its own guide; at the end you have a durable agent running in production, observed and improving.

Export a Mistral API key as MISTRAL_API_KEY: the Agents SDK reads it directly. Create one in API keys. Serving or deploying the agent also needs workflows enabled on that key.

1. Build. Install the SDK and define and run your first agent locally.
2. Add context. Give the agent capabilities: connectors and skills.
3. Test and evaluate. Evaluate the agent before deploying it, and compare versions when you change its instructions or tools.
4. Deploy. Run the same code on a worker: managed agents on Mistral infrastructure, or self-hosted agents on your own.
5. Observe. Inspect the agent's sessions in Studio traces: model calls, tool calls, and each turn.
6. Improve. Build a dataset from traces and re-run your evaluations on real conversations.