Is this project still actively maintained?

[noorbhatia]

Hey @mattt , @pcuenca

First of all thank you for building this super useful package, it has been a crucial part of the product I have been working on.
I've noticed that issues and pull requests have been accumulating without much recent activity. Is it still being actively maintained?

If bandwidth is the constraint, would you consider bringing on additional maintainers to help work through the backlog? I'd be glad to contribute with triage myself.

[pcuenca]

Hello @noorbhatia, thanks a lot for your message! Yes, we are planning to do a quick sprint in September, and see how we can evolve the project, especially within the context of the enriched provider architecture in Apple's foundation model APIs. Do you happen to have some thoughts on priority topics you'd like to see?

We'll keep you posted anyway!

[noorbhatia]

Thanks for the reply @pcuenca ! Glad to hear that there are plans for this project. I don't have any specific topics in mind, there are already couple of PRs like #181 #118 , I'd love to see some of them get merged. Besides that, support for Apple's Core AI would be great to have.

[james-333i]

@pcuenca A September sprint is good to hear. We use AnyLanguageModel in a production iOS app and have been maintaining a fork at https://github.com/james-333i/AnyLanguageModel while activity was quiet and have issued a number of PRs to move things along.

The llama.cpp backend needed the most work. My fork updates it to build against current llama.swift, enables Metal offload by default, delivers streamed tokens live instead of after generation completes, ingests prompts larger than the batch size in chunks, fixes constrained sampling ignoring the allowed-token set, and adds an assistant prefill option. I am submitting those six as PRs. Beyond them the fork adds vision input through mtmd, per-session KV cache reuse across turns, and tool calling with streaming for the Hermes JSON, Qwen 3.5 XML, and Gemma 4 formats. Those build on one another, so I will rebase and submit them once the base PRs are accepted.

On the existing queue, we have #168, #177, #180, and #181 in production and can confirm all four work well. Merging those along with our open #199, #200, and #201 would clear most of the MLX and Ollama backlog. For new work, Core AI support in the provider architecture would be our top request, followed by consistent tool calling and streaming behavior across backends.

[mattt]

Hi everyone! Apologies for the radio silence. I'm very happy to be working with @pcuenca on this project again. I'm taking time this morning to get up to speed on everything, and hope to resolve or respond to all of the open issues and PRs this week.