programatic agents with no tokens usage for workflows

[fblgit]

Description

Built this https://github.com/fblgit/agentool
It is a type of agent provider that allows input/output schematics with a router, allowing chaining and integrating with pydantic-ai somehow without much hassle.

wonder wether this could be an interesting contribution (the agentool provider) ? or wether this is solved in different ways within the framework.

thanks!

References

https://github.com/fblgit/agentool

[DouweM]

@fblgit Despite the extensive readme, it's quite unclear to me what this enables that can't be done with regular tool functions, or how Pydantic AI fits in. I found the AgenToolModel, but it doesn't seem to actually do anything with LLMs, just route the operation and return the result?

[fblgit]

the Agentool model doesnt make LLM calls, its for A2A without wasting trillions tokens with an LLM doing copy paste of the output of a tool into the input of another tool.
I believe Agents have different treatment/telemetry within pydantic-ai compared with tools, but overall flexibility?

[DouweM]

its for A2A without wasting trillions tokens with an LLM doing copy paste of the output of a tool into the input of another tool.

Can you please elaborate on the Before vs After that AgentTool makes easier/better/more performant? It's very unclear to me what you imagine a Pydantic AI integration would look like.

This sounds similar to #3093

[github-actions]

This issue is stale, and will be closed in 3 days if no reply is received.

[github-actions]

This issue is stale, and will be closed in 3 days if no reply is received.

[github-actions]

Closing this issue as it has been inactive for 10 days.