I went to the official Discourse group to see what AI is like.
Now you will see AI enabled.
We can disable part of it, but for now, we will keep it.
The main thing you will like is the proofreading it provides.
(I typed this with intentional grammar mistakes but it fixed it for me).
Discourse seems to have gone all in on the AI thing. They even have their own custom LLM (as I understand) ask.discourse.org.
We are using OpenAI DeepSeek (free) and a free OpenAI model for inline proofreading.
I used my private OpenAI+ to set up our CT AI.
For now, we are using free tools.
Discourse wants to make money, so I bet it is not free. (I did not check)
Oh.. it is just an ai chatbot. They are not charging for that.
It appears that all the features require you to plug into external models:
My understanding is that Discourse only makes money through its hosting arm.
There are tons of documentation posts about AI features:
I’m not sure if any of them would benefit this forum.
Yes… I think you are correct about the charging and how they make money.
They are also very informative about hosting your own Discourse instance, providing clear instructions on how to do it. Their hosting solution essentially operates as a “donation” model, which large corporations can and should pay to support the development.
The link you provided below is not limited to Discourse-related questions. It actually covers “anything.” When I asked it questions about AI features and plugins, it initially did not recognize what I was referring to. Then I instructed it to focus exclusively on Discourse.
We are using a free allowance from the OpenRouter website.
I am a member of that service. I created a DeepSeek model with a $0 limit,
and there was a free OpenAI feature I used for embeddings (now disabled).
I think now the balance is just right. The embeddings were a little too much.
(Proofread by ai).. I had to re-quote your quote to get the original because it fixed that too.
Interesting. I just asked it “what color is the sun” and it started out with a scientific answer and then made a funny left turn to talk about Discourse.
If you are asking about the literal color of the sun in a scientific sense, it appears white when viewed from space, and yellow from Earth due to atmospheric scattering. However, if you are referring to the color used for representing the sun in Discourse themes, icons, or emoji, this is customizable and not fixed by default in Discourse.
Discourse does not have a “sun” color variable by default. Designers can define any color they like for custom sun icons or emoji using theme customization tools.
You can customize colors for icons, emoji, or theme elements via theme editing and CSS. More on this in the guides:
For emoji or icons representing the sun, common color codes (like #FFD700 for gold/yellow) are often used, but this is up to your design preference and is not enforced by Discourse itself.
If you meant something more specific, such as a default color for a “sun” icon or feature in a particular Discourse component, please provide more details so I can give a more targeted answer.
It looks like you have another AI feature turned on:
Hmm. I would think that’s an error. It shouldn’t touch quoted text (IMHO)
Please remember to use pictures sparingly. We have a free-tier Amazon S3 bucket.
I tried to delete them but it looks like the edit window has closed. Feel free to remove them from my post. Discourse does a cleanup for resources not referenced so they will get deleted from your file system eventually.
I left one picture which was essential.. the others were not. sādhu for your activity and interest in this forum.
I recently upgraded our server and changed the AI settings to the qwen/qwen3-14b:free model. It seems zippy. It is less well-known than DeepSeek, but we don’t really need heavy AI for proofreading. If issues arise later, we will investigate a “paid” but minimal model. Probably about $1–3 per month, which is minimal.
I have changed the persona so it understands we want Pāḷi to be corrected.
Here is a sample of the persona instructions:
You are a markdown proofreader. Fix Grammar and fix Pāḷi diacritic marks for spelling on a Theravāda discourse discussion website.
You correct egregious typos and phrasing issues but keep the user’s original voice.
You do not touch quoted blocks.
If nothing needs fixing, then you will echo the text back.
Also fix unnecessary or artificial, unnatural line breaks even in quoted blocks if it looks like it was from a pdf.
We will see how it does.. We can adjust it later.
I think I have said this before somewhere else.
If you have had trouble getting the AI to proofread and waiting a long time for it to start, now things are better. It helps if we actually use a “paid” service instead of the free one. But don’t worry… so far, we are pushing seven tenths of a cent, which is up one tenth of a cent from last week.
With that, we upgraded to the Grok 4.1 paid version. Give it a try. It is quite stable now and does a good job.
