Skip to main content
An agent is the AI you want to test. In these docs the agent is a maths tutor for class 6 students. It explains one step at a time and replies in English or Hindi. You add the agent to Calibrate once. Then you write tests for it, run them, compare models, and run simulated conversations with it.

Open your agents

From the sidebar, click Agents. The list shows every agent in your workspace.
The Agents page listing the Maths homework tutor and Maths homework tutor (connected)
The Type column shows how each agent was set up. Agent means it was built in Calibrate. Connection means Calibrate calls your own agent at its URL.

Two ways to set up an agent

Click New agent. Give the agent a name, then pick one of the two options under Setup.

Build your agent in Calibrate

You write the instructions, pick the LLM, speech-to-text and text-to-speech models, and add tools. All of it lives in Calibrate.
The New agent dialog with Build your agent in Calibrate selected
See Build an agent in Calibrate.
Your agent already runs somewhere else. You give Calibrate its URL, and Calibrate sends it messages to run tests, model comparisons and simulations.
The New agent dialog with Connect your existing agent selected
See Connect your own agent.
Click Next. Then choose what your agent does:
  • Conversation: your agent has a conversation with a user. The maths tutor is this kind.
  • Basic: the agent takes an input and gives back an output.
Click Create to open the agent page.

Tabs on an agent page

Each agent page has a row of tabs. Which tabs you see depends on how the agent was set up. Click Save at the top right of the agent page to keep your changes.

Next steps

Build an agent in Calibrate

Write the prompt and pick the models

Connect your own agent

Point Calibrate at your agent’s URL