Ollama: MLX Gets a Major Upgrade and Web Search Goes Live

Transcript

Hey there, code friends! Welcome back to another episode of the Ollama podcast. I'm your host, and wow, do we have an exciting day to dive into! March 10th brought us some absolutely fantastic changes that are going to make developers' lives so much easier.

Let's jump right into the star of today's show - Daniel Hiltgen just dropped what I'm calling a game-changing MLX improvement. This pull request is massive, folks - we're talking over 5,000 lines added across 142 files. But here's why this matters for you: remember how you used to need to run cmake before building MLX code? Well, kiss those days goodbye!

Daniel implemented what's called "header vendoring" for MLX. Think of it like this - instead of your build process having to hunt around for the right header files every single time, they're now packaged right there with the code. It's like going from having to assemble your toolbox every morning to just grabbing it ready-made from the shelf. The beauty is that every time cmake runs, these headers automatically refresh, so you're always in sync with the latest MLX versions. Plus, this now works seamlessly on Windows and Linux, which is just fantastic for cross-platform development.

But wait, there's more! Parth Sareen brought us something really cool - experimental web search and web fetch routes. This is one of those features that makes you go "why didn't we have this before?" Now you can easily access web search and fetch APIs through Ollama Cloud without needing your own API keys. It's like having a research assistant built right into your workflow. And I love how they handled the edge case too - if Ollama Cloud is disabled, you get a clear, helpful error message instead of mysterious failures.

Speaking of improvements, we also saw Daniel tackle a ROCm update, bringing Linux support up to version 7.2. Now, Windows is staying on 6.2 for now, but this kind of incremental, thoughtful upgrading is exactly what you want to see in a mature project. No big bang changes that break everything - just steady, reliable progress.

Patrick Devine made model creation smoother by improving how MLX handles parameters from modelfiles. It's one of those behind-the-scenes improvements that you might not notice day-to-day, but it makes the whole experience more polished and reliable.

And here's a small but important fix that I absolutely love - Eva fixed an issue where hitting "Reset to defaults" in the app settings was accidentally disabling auto-updates. It's such a small change - literally one line - but it shows the team really cares about the user experience. These little details matter so much.

The team also squashed a Windows build issue by replacing shell wildcards with Go code for better compatibility. It's exactly the kind of fix that makes me happy as a developer - no more platform-specific headaches!

Today's Focus time! If you're working with MLX, this is your moment to update and enjoy that simplified build process. No more cmake prerequisites! If you've been wanting to experiment with web search in your Ollama projects, now's the time to check out those new experimental routes. And for everyone else, this is a perfect reminder of how incremental improvements compound over time.

What I love most about today's changes is how they show a team firing on all cylinders - major infrastructure improvements, new features, platform compatibility fixes, and attention to user experience details. That's the kind of balanced development that builds trust and momentum.

That's a wrap on today's episode! The Ollama team is clearly hitting their stride, and I'm excited to see what tomorrow brings. Keep coding, keep building, and remember - every small improvement is a step toward something amazing. Catch you next time!