Agents
Register your deployed AI application as an agent so evals can call it — the target for experiments, dataset runs, and the Conversation Simulator.
Agents
An agent is the deployed API you want to test. You register the endpoint — its URL, method, headers, auth, and body — plus a short description (or a document) of what the agent does, and evals can then call it to generate outputs and run multi-turn conversation simulations against it.
Agents are managed from the Agents item in the left sidebar.
An agent is your own HTTP endpoint that evals calls out to. This is different from API Keys, which authenticate calls made in to BrowserStack AI Evals.
What an agent holds
| Part | Description |
|---|---|
| Request spec | How to call your endpoint — Method, URL, query params, Headers, Authorization, and request Body. |
| Grounding | A short description or an uploaded document describing what the agent does. Used to keep generated test conversations on-topic. See grounding. |
| Data extraction | A mapping that tells evals where the model's answer (and any retrieved context) lives in your endpoint's JSON response. |
Where agents are used
- Experiments — pick an agent as the API Configuration so each dataset item is sent to your endpoint and the response is evaluated. See Create an Experiment.
- Dataset runs — generate a run's outputs by calling an agent.
- Scenarios — the scenario generator uses an agent's grounding to draft on-topic scenarios, and each scenario run drives the agent.
- Conversation Simulator — the agent is the system under test the simulator holds a conversation with. See Run the simulator.
In this section
Create an Agent
Register an endpoint, add grounding, map the response, and test the request.
Manage Agents
View, edit, duplicate, and delete agents.