How are AI agents configured and given a system prompt/knowledge base via the API?

AI agents are configured (in the admin, backed by the Claude API) with a role, a system prompt that serves as their instructions/knowledge base, the model, and whether they may place orders. The system prompt is where you encode what the agent knows and how it should behave; the agent then draws on live system data to answer. So tailoring an agent is largely prompt/knowledge…

AI agents are configured (in the admin, backed by the Claude API) with a role, a system prompt that serves as their instructions/knowledge base, the model, and whether they may place orders. The system prompt is where you encode what the agent knows and how it should behave; the agent then draws on live system data to answer. So tailoring an agent is largely prompt/knowledge configuration plus a few flags — you describe its job and authority, and it operates accordingly, with multiple specialized agents able to run at once.