Some questions to clarify after reading the rig documents

[qianxiaoming]

I just quickly glanced through the document and have some questions:

1. Is there no underlying functionality similar to LlmRequest? Is agent the basic way to communicate with the LLM?
2. When does the agent.prompt() exit when processing user requests? Is there a loop to automatically handle all tool calls?
3. Is the cost of building a agent very low? Is it acceptable to create a new agent for each user request?

Thanks very much!

[joshua-mo-143]

Hey @qianxiaoming !

1. You can use the CompletionModel trait - it's required for any provider to be usable in an Agent: https://docs.rs/rig-core/latest/rig/completion/request/trait.CompletionModel.html

It's implemented for practically every single provider bar 1. It also includes streaming.

2. It exits when there is only text input and no tool calls need to be executed. You can also terminate early using prompt hooks if you need to: https://docs.rs/rig-core/latest/rig/agent/trait.PromptHook.html

3. I would say so, yeah. Agents in Rig are basically just a thin wrapper around the data that you need for a request (the model, any documents you're using, etc...) so it's not like there is a huge overhead or some such. Rust is awesome like that.

Let us know if you need anything else!