# Which version of AI (Settings) for DV 4?

**URL:** <https://discourse.devontechnologies.com/t/which-version-of-ai-settings-for-dv-4/83108>\
**Category:** DEVONthink\
**Created:** [April 17, 2025, 12:51pm UTC](https://discourse.devontechnologies.com/t/which-version-of-ai-settings-for-dv-4/83108 "2025-04-17T12:51:52Z")\
**Posts on this page:** 14\
**Page:** 1

<div class="post-metadata">

**Author:** ![ophiochos](https://discourse.devontechnologies.com/letter_avatar_proxy/v4/letter/o/f04885/32.png) [@ophiochos](https://discourse.devontechnologies.com/u/ophiochos)\
**Post date:** [April 17, 2025, 12:51pm UTC](https://discourse.devontechnologies.com/t/which-version-of-ai-settings-for-dv-4/83108/1 "2025-04-17T12:51:52Z")

</div>

Hi folks,

I’ve not used AI much but the beta of DV 4 made me think it could be useful so I dug out my chatGPT free account details. DV wants to know which version I am using, in the settings, but ChatGPT told me it is using ‘GPT-4- turbo’ (not listed as an option).  
It also promised they are planning to simplify the naming methods (which is a great relief, looking at the list of random suffixes they use).

Does it matter which one I put in DV settings?

Cheers,

J

---

<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:** [April 17, 2025, 1:47pm UTC](https://discourse.devontechnologies.com/t/which-version-of-ai-settings-for-dv-4/83108/2 "2025-04-17T13:47:57Z")

</div>

First of all, you need an API key to use a commercial AI service. Just an account to chat with ChatGPT is not enough. And yes, this likely will incur extra costs for you.

There are specific models we make available, based on capability and cost.

Also, check out the **Getting Started \> AI Explained** section in the Help.

---

<div class="post-metadata">

**Author:** ![Nielle](https://discourse.devontechnologies.com/letter_avatar_proxy/v4/letter/n/8c91f0/32.png) [@Nielle](https://discourse.devontechnologies.com/u/Nielle)\
**Post date:** [April 17, 2025, 2:37pm UTC](https://discourse.devontechnologies.com/t/which-version-of-ai-settings-for-dv-4/83108/4 "2025-04-17T14:37:03Z")

</div>

Is there a way or plan to use Apple Intelligence in DT? I believe it’s a special fork of ChatGPT and doesn’t require any API to work, no?

---

<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:** [April 17, 2025, 3:00pm UTC](https://discourse.devontechnologies.com/t/which-version-of-ai-settings-for-dv-4/83108/5 "2025-04-17T15:00:16Z")

</div>

No. Apple Intelligence is not an accessible API call to ChatGPT.

---

<div class="post-metadata">

**Author:** ![RobH](https://discourse.devontechnologies.com/user_avatar/discourse.devontechnologies.com/robh/32/7787_2.png) [@RobH](https://discourse.devontechnologies.com/u/RobH)\
**Post date:** [April 17, 2025, 3:09pm UTC](https://discourse.devontechnologies.com/t/which-version-of-ai-settings-for-dv-4/83108/6 "2025-04-17T15:09:11Z")

</div>

I’m curious which part of Apple Intelligence you were wanting to use in DT? I’ve found Apple’s implementation to be… let’s just call it underwhelming. Is there anything Apple’s AI can do that can’t be done better with any other AI? I’m not trying to be crass, I’m honestly curious.

---

<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:** [April 17, 2025, 3:15pm UTC](https://discourse.devontechnologies.com/t/which-version-of-ai-settings-for-dv-4/83108/7 "2025-04-17T15:15:09Z")

</div>

> [@RobH](#):
>
> I’ve found Apple’s implementation to be… let’s just call it underwhelming.

That’s generous 😉

---

<div class="post-metadata">

**Author:** ![kewms](https://discourse.devontechnologies.com/letter_avatar_proxy/v4/letter/k/85f322/32.png) [@kewms](https://discourse.devontechnologies.com/u/kewms)\
**Post date:** [April 17, 2025, 5:22pm UTC](https://discourse.devontechnologies.com/t/which-version-of-ai-settings-for-dv-4/83108/8 "2025-04-17T17:22:38Z")

</div>

Apple Intelligence is much more transparent about when it is (and isn’t) transferring your data to another computer. Its “local-first” model is part of _why_ it has underwhelming performance, I suspect.

Whether you care about that probably depends on the nature of your data.

---

<div class="post-metadata">

**Author:** ![RobH](https://discourse.devontechnologies.com/user_avatar/discourse.devontechnologies.com/robh/32/7787_2.png) [@RobH](https://discourse.devontechnologies.com/u/RobH)\
**Post date:** [April 17, 2025, 7:41pm UTC](https://discourse.devontechnologies.com/t/which-version-of-ai-settings-for-dv-4/83108/9 "2025-04-17T19:41:26Z")

</div>

Okay, but that doesn’t really answer the question. You can use a local LLM if you want privacy, so that’s not a reason to use Apple’s AI.

I’m genuinely curious why one would want that as their main AI. Maybe I missed something on what it does, and if so, I’d like to know.

Like many, I eagerly waited for Apple’s AI to be released, thinking that it was going to provide us the ability to use AI without worrying about privacy. Unfortunately, it is so much vaporware it’s not even worth considering, assuming I haven’t overlooked that killer feature of it.

---

<div class="post-metadata">

**Author:** ![ubergene](https://discourse.devontechnologies.com/user_avatar/discourse.devontechnologies.com/ubergene/32/19186_2.png) [@ubergene](https://discourse.devontechnologies.com/u/ubergene)\
**Post date:** [April 18, 2025, 2:01am UTC](https://discourse.devontechnologies.com/t/which-version-of-ai-settings-for-dv-4/83108/10 "2025-04-18T02:01:37Z")

</div>

With DT4 it seems there are AI functionalities one can use without having to subscribe to (another) AI model - am I correct? In other words, does DT4 itself incorporate a solid LLM-type sub-routine?

---

<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:** [April 18, 2025, 2:03am UTC](https://discourse.devontechnologies.com/t/which-version-of-ai-settings-for-dv-4/83108/11 "2025-04-18T02:03:17Z")

</div>

No, there is no LLM built into DEVONthink 4. Read the **Getting Started \> AI Explained** section about how access to external AI is integrated.

---

<div class="post-metadata">

**Author:** ![chrillek](https://discourse.devontechnologies.com/user_avatar/discourse.devontechnologies.com/chrillek/32/29340_2.png) [@chrillek](https://discourse.devontechnologies.com/u/chrillek)\
**Post date:** [April 18, 2025, 9:53am UTC](https://discourse.devontechnologies.com/t/which-version-of-ai-settings-for-dv-4/83108/12 "2025-04-18T09:53:39Z")

</div>

> [@RobH](#):
>
> You can use a local LLM if you want privacy

With the trade-off of underwhelming performance, according to the posts here.

> [@RobH](#):
>
> I’m genuinely curious why one would want that as their main AI

AFAICT, Apple’s AI does some basic stuff like image classification locally. I don’t know if translation is considered “AI”, but that happens locally, too.

The privacy vs performance issue is very much a having and eating a cake one, imo. One _can_ do basic things on a local machine, apparently – privacy. But for anything interesting, a remote service must be used – no privacy.

Expecting to run a large, helpful LLM on a comparatively small machine is delusion.

---

<div class="post-metadata">

**Author:** ![Nielle](https://discourse.devontechnologies.com/letter_avatar_proxy/v4/letter/n/8c91f0/32.png) [@Nielle](https://discourse.devontechnologies.com/u/Nielle)\
**Post date:** [April 19, 2025, 4:27pm UTC](https://discourse.devontechnologies.com/t/which-version-of-ai-settings-for-dv-4/83108/13 "2025-04-19T16:27:03Z")

</div>

Well since Apple has applie intelligence already on Mac, it’s a bit annoying to download a local LLM. And since I have a MacBook Air 2020, local LLM doesn’t perform well where I have no issues with Apple intelligence.

Let’s see how that will evolve in the future.

---

<div class="post-metadata">

**Author:** ![anonny](https://discourse.devontechnologies.com/letter_avatar_proxy/v4/letter/a/a8b319/32.png) [@anonny](https://discourse.devontechnologies.com/u/anonny)\
**Post date:** [April 20, 2025, 5:24am UTC](https://discourse.devontechnologies.com/t/which-version-of-ai-settings-for-dv-4/83108/14 "2025-04-20T05:24:33Z")

</div>

You can’t trust any AI to tell you what version they are. Even none openai models will say they’re chatgpt. To control which version you use, there is an up and down arrow next to the chat dialogue box.  
Also, almost all interaction with AIs need an API key.

 ![Screenshot 2025-04-20 at 12.23.18](https://devontech-discourse.s3.dualstack.us-east-1.amazonaws.com/uploads/original/3X/4/9/49c41bcbc13a8c6acec21b1cdecaa2b46694c7c5.png)

---

<div class="post-metadata">

**Author:** ![ophiochos](https://discourse.devontechnologies.com/letter_avatar_proxy/v4/letter/o/f04885/32.png) [@ophiochos](https://discourse.devontechnologies.com/u/ophiochos)\
**Post date:** [April 23, 2025, 5:00pm UTC](https://discourse.devontechnologies.com/t/which-version-of-ai-settings-for-dv-4/83108/15 "2025-04-23T17:00:54Z")

</div>

lol we have gone full circle. I don’t know which of those listed in the screenshot is the one I have access to. But Bluefrog’s already sent me in the right direction so I’ll have a look there.