An agent: a reason, act, observe loop with a prompt, a script, operations, functions and other agents around it, and a model chosen per agent from Anthropic, OpenAI or Bedrock

What an agent is

A model in a loop. The configuration is the agent.

A ReAct loop: reason, call a tool, read the result, repeat. Each turn runs within limits on time, model rounds, tokens and repeated calls.

Everything around the loop is configuration, and lives in Mindset.

2 of 11
Speaker notes

The prompt is five fields: purpose, personality, background, policy rules and output formatting, up to 2,000 characters each. The model is chosen per agent from the providers your org has keys for: Anthropic, OpenAI or AWS Bedrock. The model is a swappable call, so changing it is a new version, not a rewrite. The script, if there is one, holds the agent to phases. Resources are what it may call: operations, functions, widgets and other agents. A turn that reaches a limit stops where it is and records step_limit, deadline, spend_limit or repetition. Mindset enforces the resource list whatever the prompt says.