Comment by mchiang
4 days ago
one of the maintainers for Ollama. I don't see it as a pivot. We are all developers ourselves, and we use it.
In fact, there are many self-made prototypes before this from different individuals. We were hooked, so we built it for ourselves.
Ollama is made for developers, and our focus in continually improving Ollama's capabilities.
Congratulations on launching the front-end, but I don't see how it can be made for developers and not have a Linux version.
I've never used a single linux GUI app in my 15 years of developing software. No company I've worked for even gives out linux laptops.
Its very strange, but they do have a Linux client that they refuse to mention in their blog post. I have no idea if this is a simple slip-up or if it was for some reason intentional.
https://ollama.com/download/linux
This link is for the existing cli version, not the new gui app.
Whoah, are you telling me that there are devs on Linux who use anything else than a tiled WM? CLI or GTFO /s
I just updated and a bit annoying by default gemma3:4b was selected that I don't have on my local. I guess would be nicer to default to one of the models that are present.
It was nice it started downloading it but also there was no indication I don't have that model before hand until I opened drop-down to see download buttons.
But of course nice job guys.
Thanks for the kind words. Sorry about that, we are working out some of the initial experience for Ollama.
The need to start a chat for a model that is not currently downloaded in order to initiate a download confused me for a minute the first time I tried it out. A more intuitive approach (which is the first thing I tried to do before I figured it out) might be to make the download icons in the model list clickable, to initiate a download. Then you could display a download progress bar in the list, and when models have been downloaded show a little "info" icon that is also clickable to display the model card, surface other model specific options, and enable deletion. Love the new UI, kudos!
Big thanks to you and your team for this. My first time tying offline models, will the github cli use the same models by default (MacOS)?
Congratulations on the next release.
I really like using ollama as a backend to OpenWebUI.
I don't have any windows machines and I don't work primarily on macos, but I understand that's where all the paying developers are, in theory.
Did y'all consider a partnership with one of the existing UI and bundle that, similar to duckdb approach?
I’m just curious because I don’t use Ollama and have some spare vram: How do you use it and what models do you use?
Thanks for including it. Ollama is very good at what it does. Including the feature is showing mindful growth in helping ollama be that skateboard, scooter, car, etc that the developer needs for LLM at that time. Making it appeal to casual/hobbyist is the right approach.
PS totally running windows here and using kesor/ollama-proxy if I need to make it externally available.
do you know why ollama hasn't updated its models in over a month while many fantastic models have been released in that time, most recently GLM 4.5? It's forcing me to use LM Studio which I for whatever reason absolutely do not prefer.
thank you guys for all your work on it, regardless
You know that if you go to hugging face and find a gguf page, you can click on Deploy and select ollama. It comes with “run” but whatever—just change to pull. Has a jacked name, but works.
Also, if you search on ollama’s models, you’ll see user ones that you can download too
GLM 4.5 has a new/modified architecture. From what I understand, MLX was really one of the only frameworks that had support for it as of yesterday. LM Studio supports MLX as one backend. Everyone else was/is still developing support for it.
Ollama has the new 235B and 30B Qwen3 models from this week, so it’s not as if they have done nothing for a month.
ah, that explains why all the GLM quants are MLX models
We work closely with majority of research labs / model creates directly. Most of the times we will support models on release day. There are sometimes where the release window for major models are fairly close - and we just have to elect to support models where we believe will better support a majority of users.
Nothing out of spite, and purely limited by the amount of effort required to support these models.
We are hopeful too -- where users can technically add models to Ollama directly. Although there is definitely some learning curve.
Would love to add models direclty. And don't worry, we will figure it out!
just so you know, you can grab any gguf from huggingface and specify the quant like this:
qwen3 was updated less than a day ago: https://ollama.com/library/qwen3
Question since you are here, how long before tool-calling is enabled for Gemma3 models?
Seems that Google intend it to be that way - https://ai.google.dev/gemma/docs/capabilities/function-calli... . I suppose they are saying that the model is good enough that if you put the tool call format in prompt it should be able to handle any formats.
I use PetrosStav/gemma3-tools and it seems that it only works half of the time - the rest the model call the tool but it doesn't get properly parsed by Ollama.
unfortunately, I don't think gemma 3 supports tool calling well. It's not trained into the model, and the 'support' for tool calling is post model training.
We are working with Google, and trying to give the feedback on improving tool calling capabilities for future Gemma models. Fingers crossed!
2 replies →
You can do “tool calling” via Gemma3. The issue is that it all needs to be stuck in the user prompt as there’s no system prompt
While you are here: what's the state of external contributing for the project? I see literally hundreds of open PRs, and it's a bit discouraging, tbh.
Are there any plans to improve observability toolset for developers? There is myriad of various AI chat apps, and there is no clear reason why another one from Ollama would be better. But Ollama is uniquely positioned to provide the best observability experience to its users because it owns the whole server stack, any other observability tool (eg Langfuse) may only treat it as a yet another API black box.
Does the new app make it easier for users to expose the Ollama daemon on the network (and mdns discovery )? It’s still trickier than needed for Home Assistant users to get started with Ollama (which tends to run on a different machine).
In the app settings now, there is a toggle for "Expose Ollama to the network" - it allows for other devices or services on the network to access Ollama.
There's a simple toggle for juet that.
I think welcoming another stream of users, in addition to develoeprs is a good idea.
Lots of people trying to being, and many with Ollama, and helping to create beginners is never a bad thing with tech.
This caught me out yesterday. I was trying to move models onto external disk, and it seems to require re-installation? but there was no sign of the simple CLI option that was previously presented and I gave up.
As a developer feature request, it would be great if ollama could support more than one location at once, so that it is possible to keep a couple models 'live' but have the option to plug in an external disk with extra models being picked up auto-magically based on the ollama_models path please. Or maybe the server could present a simple html interface next to the API endpoint?
And just to say thanks for making these models easily accessible. I am agAInst AI generally, but it is nice to be able to have a play with these models locally. I havent found one that covers Zig, but appreciate the steady stream of new models to try. Thanks.
I just symbolically link the default model directory to a fast and cheap external drive. I agree that it would be nice to support multiple model directories.
I think welcoming another stream of users, in addition to develoeprs is a good idea.
Lots of people trying to being, and many with Ollama, and helping to create beginners is never a bad thing with tech.
Many things can be for both developers and end-users. Developers can use the API directly, end users, have more choices.
How about include MCP option?