# Using DEVONthink 4.4 with oMLX

**URL:** <https://discourse.devontechnologies.com/t/using-devonthink-4-4-with-omlx/87529>\
**Category:** Artificial Intelligence\
**Created:** [August 27, 2026, 4:49pm UTC](https://discourse.devontechnologies.com/t/using-devonthink-4-4-with-omlx/87529 "2026-08-27T16:49:50Z")\
**Posts on this page:** 15\
**Page:** 1

<div class="post-metadata">

**Author:** ![Tekl](https://discourse.devontechnologies.com/user_avatar/discourse.devontechnologies.com/tekl/32/5508_2.png) [@Tekl](https://discourse.devontechnologies.com/u/Tekl)\
**Post date:** [August 27, 2026, 4:49pm UTC](https://discourse.devontechnologies.com/t/using-devonthink-4-4-with-omlx/87529/1 "2026-08-27T16:49:50Z")

</div>

I’m currently trying to optimize the performance of my local AI models. I’ve already experimented a lot with LM Studio and Ollama, and I’ve now come across oMLX, which offers more options for speculative decoding. I’m hoping this will speed things up. The problem is this: I’ve entered the local server’s URL in DEVONthink as the OpenAI API URL: [http://localhost:8000/v1/chat/completion](http://localhost:8000/v1/chat/completion).

However, this means that the selection of available models isn’t immediately accessible in DEVONthink; instead, oMLX loads each model into memory and then unloads it (once the memory limit is reached). I have over 30 models I’d like to try out. Consequently, it takes forever—with high CPU usage—before I can select a model in DEVONthink.

Of course, I’ve already asked ChatGPT what’s going on. The theory is that DEVONthink does more than just call GET /v1/models, which is why oMLX is forced to load each model. Is that correct? Does anyone know their way around oMLX and have any ideas? The AI just suggests to use fewer models. But that makes comparing different models quite cumbersome.

---

<div class="post-metadata">

**Author:** ![cgrunenberg](https://discourse.devontechnologies.com/user_avatar/discourse.devontechnologies.com/cgrunenberg/32/7172_2.png) [@cgrunenberg](https://discourse.devontechnologies.com/u/cgrunenberg)\
**Post date:** [August 27, 2026, 4:55pm UTC](https://discourse.devontechnologies.com/t/using-devonthink-4-4-with-omlx/87529/2 "2026-08-27T16:55:01Z")

</div>

> [@Tekl](#):
>
> The problem is this: I’ve entered the local server’s URL in DEVONthink as the OpenAI API URL: [http://localhost:8000/v1/chat/completion](http://localhost:8000/v1/chat/completion).

If the service doesn’t provide the models’ capabilities (e.g. LM Studio, Ollama or OpenRouter do this), then DEVONthink probes each model once to check whether it supports chat completions, tool calls and/or vision. A future release might skip this in case of local, unknown services & models but then the possibilities would be limited.

---

<div class="post-metadata">

**Author:** ![Tekl](https://discourse.devontechnologies.com/user_avatar/discourse.devontechnologies.com/tekl/32/5508_2.png) [@Tekl](https://discourse.devontechnologies.com/u/Tekl)\
**Post date:** [August 27, 2026, 7:41pm UTC](https://discourse.devontechnologies.com/t/using-devonthink-4-4-with-omlx/87529/3 "2026-08-27T19:41:20Z")

</div>

Thanks for the explanation. So oMLX should provide this information.

---

<div class="post-metadata">

**Author:** ![BLUEFROG](https://discourse.devontechnologies.com/user_avatar/discourse.devontechnologies.com/bluefrog/32/135_2.png) [@BLUEFROG](https://discourse.devontechnologies.com/u/BLUEFROG)\
**Post date:** [August 27, 2026, 8:21pm UTC](https://discourse.devontechnologies.com/t/using-devonthink-4-4-with-omlx/87529/4 "2026-08-27T20:21:19Z")

</div>

oMLX is still quite buggy, by the way.

---

<div class="post-metadata">

**Author:** ![cgrunenberg](https://discourse.devontechnologies.com/user_avatar/discourse.devontechnologies.com/cgrunenberg/32/7172_2.png) [@cgrunenberg](https://discourse.devontechnologies.com/u/cgrunenberg)\
**Post date:** [August 28, 2026, 5:50am UTC](https://discourse.devontechnologies.com/t/using-devonthink-4-4-with-omlx/87529/5 "2026-08-28T05:50:21Z")

</div>

We will check this.

---

<div class="post-metadata">

**Author:** ![cgrunenberg](https://discourse.devontechnologies.com/user_avatar/discourse.devontechnologies.com/cgrunenberg/32/7172_2.png) [@cgrunenberg](https://discourse.devontechnologies.com/u/cgrunenberg)\
**Post date:** [August 28, 2026, 10:23am UTC](https://discourse.devontechnologies.com/t/using-devonthink-4-4-with-omlx/87529/6 "2026-08-28T10:23:03Z")

</div>

> [@Tekl](#):
>
> I’ve already experimented a lot with LM Studio and Ollama, and I’ve now come across oMLX, which offers more options for speculative decoding. I’m hoping this will speed things up.

At least on my machine (M1 Ultra) using the default settings and a context window of 32k Ollama is approx. 1.6 times faster than oMLX when using Gemma 4:26b and 2.8-3.5 times faster when using Qwen3.8:27b. Latest app versions in both cases and 4-bit MLX models, tested using DEVONthink (_Data \> Tags \> Add chat suggestions to documents_).

And oMLX does indeed not support loading of model infos:

`CHAT: ERROR: Loading of model info failed: Invalid HTTP status 404`

---

<div class="post-metadata">

**Author:** ![ALHOFF](https://discourse.devontechnologies.com/letter_avatar_proxy/v4/letter/a/c6cbf5/32.png) [@ALHOFF](https://discourse.devontechnologies.com/u/ALHOFF)\
**Post date:** [August 28, 2026, 4:37pm UTC](https://discourse.devontechnologies.com/t/using-devonthink-4-4-with-omlx/87529/7 "2026-08-28T16:37:57Z")

</div>

Start your model before starting DT and connecting it. I’m running MTPLX, and it connected immediately in DEVONthink and has run smoothly since.

---

<div class="post-metadata">

**Author:** ![Tekl](https://discourse.devontechnologies.com/user_avatar/discourse.devontechnologies.com/tekl/32/5508_2.png) [@Tekl](https://discourse.devontechnologies.com/u/Tekl)\
**Post date:** [August 28, 2026, 5:23pm UTC](https://discourse.devontechnologies.com/t/using-devonthink-4-4-with-omlx/87529/8 "2026-08-28T17:23:48Z")

</div>

Thanks for testing. Very interesting. My hope was to get beyond Ollama by using MTP with a draft model. Maybe Ollama is doing this out of the box and I should give it a chance again. I’ve ditched Ollama because the models are not interchangeable with different inference engines. Due to slow internet it’s not easy for me to try all those fat downloads. 😜

---

<div class="post-metadata">

**Author:** ![Tekl](https://discourse.devontechnologies.com/user_avatar/discourse.devontechnologies.com/tekl/32/5508_2.png) [@Tekl](https://discourse.devontechnologies.com/u/Tekl)\
**Post date:** [August 28, 2026, 5:25pm UTC](https://discourse.devontechnologies.com/t/using-devonthink-4-4-with-omlx/87529/9 "2026-08-28T17:25:00Z")

</div>

I already pinned a model. But how do you skip loading the model list in DEVONthink? Do add the model to the API url?

---

<div class="post-metadata">

**Author:** ![cgrunenberg](https://discourse.devontechnologies.com/user_avatar/discourse.devontechnologies.com/cgrunenberg/32/7172_2.png) [@cgrunenberg](https://discourse.devontechnologies.com/u/cgrunenberg)\
**Post date:** [August 28, 2026, 5:37pm UTC](https://discourse.devontechnologies.com/t/using-devonthink-4-4-with-omlx/87529/10 "2026-08-28T17:37:19Z")

</div>

> [@Tekl](#):
>
> But how do you skip loading the model list in DEVONthink?

That’s not possible currently.

---

<div class="post-metadata">

**Author:** ![ALHOFF](https://discourse.devontechnologies.com/letter_avatar_proxy/v4/letter/a/c6cbf5/32.png) [@ALHOFF](https://discourse.devontechnologies.com/u/ALHOFF)\
**Post date:** [August 28, 2026, 7:02pm UTC](https://discourse.devontechnologies.com/t/using-devonthink-4-4-with-omlx/87529/11 "2026-08-28T19:02:42Z")

</div>

Have oMLX running and model loaded. Then Open DT.  
I’m currently doing this with MTPLX but also have run oMLX (but not yet with DT).  
Nonetheless, this is a DT finding the model, not an issue at the model server side.

---

<div class="post-metadata">

**Author:** ![Tekl](https://discourse.devontechnologies.com/user_avatar/discourse.devontechnologies.com/tekl/32/5508_2.png) [@Tekl](https://discourse.devontechnologies.com/u/Tekl)\
**Post date:** [August 29, 2026, 5:06am UTC](https://discourse.devontechnologies.com/t/using-devonthink-4-4-with-omlx/87529/12 "2026-08-29T05:06:36Z")

</div>

You don’t need to select the model in the AI configuration? And do you have multiple Models (\>30) available in oMLX? With only 3 models on your disk, my issue wouldn’t be a problem.

---

<div class="post-metadata">

**Author:** ![Mkcmobile](https://discourse.devontechnologies.com/letter_avatar_proxy/v4/letter/m/c37758/32.png) [@Mkcmobile](https://discourse.devontechnologies.com/u/Mkcmobile)\
**Post date:** [August 29, 2026, 1:36pm UTC](https://discourse.devontechnologies.com/t/using-devonthink-4-4-with-omlx/87529/13 "2026-08-29T13:36:48Z")

</div>

I’ve not tried so many models at once, and as you say, with only handful at once (currently 5) it’s not a problem. Is the reason you have so many that you want to test them with DT? I use Pi and oMLX for research and I’ll find I quickly end up narrowing down to a shorter list of models if I think about testing new ones. Say one thought, I don’t know how you are downloading the models, but I use LM Studio to determine if I want something new, then download there. I point oMLX to the LM Studio location as well as its own. You could probably easily segregate your models, without losing any, while you test a subset of the 30.

---

<div class="post-metadata">

**Author:** ![BLUEFROG](https://discourse.devontechnologies.com/user_avatar/discourse.devontechnologies.com/bluefrog/32/135_2.png) [@BLUEFROG](https://discourse.devontechnologies.com/u/BLUEFROG)\
**Post date:** [August 29, 2026, 2:35pm UTC](https://discourse.devontechnologies.com/t/using-devonthink-4-4-with-omlx/87529/14 "2026-08-29T14:35:07Z")

</div>

Yes, why do you have 30 models installed

---

<div class="post-metadata">

**Author:** ![Tekl](https://discourse.devontechnologies.com/user_avatar/discourse.devontechnologies.com/tekl/32/5508_2.png) [@Tekl](https://discourse.devontechnologies.com/u/Tekl)\
**Post date:** [October 16, 2026, 7:54pm UTC](https://discourse.devontechnologies.com/t/using-devonthink-4-4-with-omlx/87529/15 "2026-10-16T19:54:42Z")

</div>

Well. I’m a bit late. I’ve got lots of Models to test them, so I can select the best one for my machine. For example there are a lot of quantizations and optimizations for Qwen3.8 27B. And even the Runtime can make a huge difference. I recently switched to MLX Server with its own Qwen Distill and I get 80 TOK/s instead of 50. Most models are deleted benchmarking, but I’m downloading faster than I have time to benchmark. 😁