Test runs

A test run puts your real agent in front of a fake event. Your model, your prompt, and your wiring all run for real, but every tool call the agent makes is answered with made-up data instead of going to your accounts. Nothing is sent, nothing is written, and nothing is deleted. That lets you answer the question you actually have before an agent goes live: given a message like this, what would it do?

​

Start a test run

You can also reach a test run from the Setup tab, where it is the last step. Click Exit Testing to go back to the chat, or Talk to your agent once a run finishes. Testing does not wait for setup to be complete. Because the tool calls are answered by the simulation, an agent with nothing connected yet still runs end to end.

​

Staged situations

The situation is the event the agent wakes up on. Gmail, Slack, WhatsApp, and Telegram triggers ship written situations covering the cases those channels really see. Every other trigger gets a sample built from your own graph. Situations render in the shape they would really arrive in, so a WhatsApp message looks like WhatsApp and a Gmail message looks like an email. You can edit the staged message in place. The edit is real: the agent receives the payload as you rewrote it, so you can push it toward the awkward case you are worried about.

​

Several runs per agent

Use New run to keep more than one situation on hand, and the pencil to rename them. A run is a saved test, so an agent can carry a small suite: the happy path, the refund request, the person who asks to be left alone. Runs are stored with the workflow, which means they travel with it. Publish the workflow as a template and whoever takes a copy gets your test suite along with the agent.

​

What is simulated, and what is not

The Test Run screen leads with a line saying the whole thing is a simulation, and every answered tool call carries a simulated pill. Outbound messages show as what would have been sent rather than as sent. Real: the agent, its model, its prompt, its tools, and the decisions it makes. Simulated: every answer coming back from a tool, and every outbound message.

​

Share a finished run

Click Share this run at the end of a finished run to mint a read-only page at https://noclick.com/r/{id} and copy the link. Anyone with the link can open it, with no account needed. The page is a static snapshot of the run you just watched, headed by the same line saying the data is simulated. Nothing executes when it is opened, nothing bills, and later edits to your workflow never leak into it. Only the workflow owner can create one.

​

What it costs

The agent’s own model usage is billed the way any agent run is billed. The simulated world’s replies run on NoClick’s models and are billed separately under the Agent Testing category, so you can tell rehearsal spend apart from your agent’s spend in Usage.

​

Next steps