Testing an agent
Try your agent live in the builder, read its trace, and iterate before deploying.
You can run an agent without leaving the builder. The Test panel lets you talk to the current draft, watch it work in real time, and inspect exactly what it did — so you can refine it before deploying.
The test panel
Click Test to open the panel. Type a message — for a chat agent, anything; for an object-input agent, the input it expects — and send it. The agent runs and its response streams back live, just as it will in production.
As it works, a collapsible trace shows each tool call and the token count, so you can see not just the answer but how the agent got there. If a tool call fails, the error shows up in the trace too.
Test conversations are throwaway
Anything you do in the test panel is ephemeral — it isn't saved as a run and won't appear in Observability. Test as much as you like; nothing you try here affects your live agent or its history.
If the agent has conversation history turned on, your earlier test messages are replayed as context, so you can check how it handles a multi-turn exchange.
Iterate
Testing is meant to be a tight loop:
- Send a representative input.
- Read the response and the trace.
- Adjust the instructions, model, parameters, or tools.
- Test again.
Because edits auto-save to the draft, each test runs your latest changes.
Deploy when it's ready
Once the agent behaves the way you want, Deploy it to freeze the draft into a numbered version and publish it live. Deploying runs a quick validation and blocks on anything that would stop the agent from running. After it's live, your team can run it and other agents and workflows can use it as a tool.