An agent engine powers an agent by connecting its profile and base behavior to language models and tools. It provides the implementation that turns user input into responses. section 7.3 section 7.4
A practical agent engine typically:
- Subclasses a base agent and implements response methods. section 7.4
- Connects to a model such as OpenAI, while allowing other model or tool platforms. section 7.4
- Discovers actions through a plugin folder and supports tool specifications and function calling. section 7.4 section 7.5
- Executes multiple tool calls and sends their results back to the model for a response. section 7.5
OpenAI-powered response
An OpenAI agent appends the user message, calls the chat-completions API with its model and messages, stores the returned content, and returns it. With tools enabled, it instead supplies tool specifications, executes requested functions, and makes a follow-up model call. section 7.4 section 7.5
