Repo Roundup #6

Of course this week was pretty much dominated with AI announcements so much of my explorations around the Internet were about that.

changesets

I’m trying this on some of my projects. It’s an opinionated toolset for creating and managing changesets for repositories with automatic versioning and CHANGELOG.md generation.

tsify

This is a library that can generate TypeScript definitions from Rust code. I used it when I was working on my TypeScript wrapper for the OpenAI Harmony renderer. It automatically generates TypeScript types for the Rust structs defined in OpenAI’s Harmony renderer Rust source code. It was very easy to get it working with wasm-bindgen, even though I don’t know much about Rust or the wasm-bindgen library.

harmony

OpenAI’s new GPT-OSS models use a new message format called Harmony. It is essentially required to use the Harmony format; otherwise, the model will not work as intended. This is OpenAI’s official renderer for the Harmony format to help create appropriate messages. The renderer is written in Rust and compiles down to Python bindings as well.

responses.js

This is a small Express.js server that sits in front of an OpenAI-compatible inference API, and can quickly rewrite responses to use OpenAI’s new Responses API. I briefly used this because Ollama does have an OpenAI-compatible API built into it, but it doesn’t support the Responses API. As I was using Ollama to serve the new gpt-oss model I was attempting to use this to make it more compatible with some tools I was trying to use.

gpt-oss

OpenAI’s new open-sourced GPT models. I’ve been enjoying playing with them and seeing where it would make sense to use them in my projects and where it does not. So far it has been difficult to truly evaluate them given inconsistent implementations across providers and tools of the new Harmony message format.