Build a simple agent
This guide walks you through building a minimal working agent end to end. Once the base agent is set up, you can explore how to extend it with additional features, each linked to the relevant documentation.
Create a model
Choosing the right provider and model depends on your agent’s purpose and the type of interaction you want to support.
-
In the Cockpit, open the Model tool.
-
Select Create.
-
Select a provider.
-
Set Input Type to Text and Output Type to Text. This ensures the model is an LLM and not an embedding model.
-
Fill in the remaining fields and select Create.
-
Add your API key or other required credentials, then select Save.
Now that you have a model, you can create an agent that uses it.
Create an agent
-
In the Cockpit, open the Agent tool.
-
Select Create.
-
Select your model, fill in the remaining required fields, and select Create.
For an agent to work it needs an instruction. Navigate to the Instructions tab to set one up:
-
Open the Instructions tab.
-
Create a custom instruction or select one of the provided examples.
-
Select Save.
Test your agent
You can test the agent directly in the Playground tab.
Default behavior
A minimal agent created with the steps above has the following default properties:
-
The temperature, if supported by the model, is set to 1 (Balanced).
-
The context window retains the last 3 previous messages.
-
No tools are configured.
-
No guardrails are configured.
-
No MCP connections are configured.
-
No role-based access restrictions are applied to the agent.
-
Every request is automatically logged and visible in the Agent Trace.