AI & MLLLMOpenAI

Keeping Agent Framework References Alongside Custom Voice Work

Building an application around a model raises questions that a single request does not answer. Where does state live? How are tools described? What happens when a workflow moves between responsibilities?

Two repositories on my GitHub account preserve upstream references for those questions: OpenAI's Python Agents SDK and Realtime Agents examples.

Frameworks and application responsibilities

An orchestration framework can provide structure around model requests, tools, and execution flow. A Realtime example addresses another set of concerns around live interaction.

The useful comparison is with a specific application requirement. For my voice projects, that might be how an example represents a conversation's state or separates a tool request from the code carrying it out.

Those questions also help identify what remains application-specific. Telephone routing, recipient context, delivery evidence, and billing rules do not become correct merely because a model sits inside an agent framework.

Why preserve a fork

A fork gives an upstream project a visible place in a working collection. It can serve as a reference point for reading examples or developing a later adaptation while preserving the original project's identity.

The forks were created in January 2026. That is a date for the repository relationship, not proof that I implemented the framework in a product. No matching local checkout was identified in the surveyed project roots.

The next useful comparison

I would choose one small example and compare its responsibilities with the custom Realtime caller and deterministic reminder service. That would make it easier to decide which framework patterns help and which product boundaries still need explicit code.

This is a learning/reference entry in the history, with the upstream authorship kept visible.

Repositories