We’re thrilled to announce that our hosted customers can now power every Discourse AI feature using our own hosted, open-weights LLM[1], pre-configured and included free of charge with your hosting service.
This means every AI feature is available to Starter, Pro, Business, and Enterprise customers without needing any third-party LLM provider.
Enabling the hosted LLM
This should be already available and enabled on your site under the LLM configuration page:
By default, “CDCK Hosted Small LLM” will be selected as your default LLM, and it should also be available for selection by any persona on your site.
Usage and limits
Discourse measures hosted LLM usage through a credit system. Each request or response token consumes 1 credit. Credits are proportional to your hosting tier and are reset daily.
✱ Some features such as AI Spam detection will not be counted towards your credits
Credit limits per tier can be found on the pricing page on our website. After you reach your daily limit, any AI features that rely on the LLM will pause until credits reset the next day.
If you’re running out of credits quickly, consider using the LLM quota system to set per-user or per-group restrictions. Upgrading to a higher tier is another option if you need additional capacity.
You can review your site’s AI usage at any time from the AI usage page in the admin panel.
For more details on what to do when credits are depleted, see:
So is the AI Summarize feature only available against the master post for a topic? It would be valuable to have this for a long and complex reply within a thread.
The Starter tier has 15K daily “AI Credits”. What does one credit equal? Is that a single token in a reply? It would be useful if you would show at the bottom of each AI summary how many credits it used.
Summarizing the entire topic is useful, but that’s not the case I discussed. In a topic with 150 replies, I am reading one specific reply that is long and complicated. I want that one reply summarized by AI. We have this now on Twitter and it is a fantastic and useful feature.
If a single AI Summary takes 1500 AI credits, and we only get 15K credits/day, that means we only need 10 people to request an AI summary and we have used up everything. That makes the feature not very useful.
So if I jump from $20/month to $100/month, now I get 30K AI credits per day, and only 20 users using the feature exhaust my credits? I think this is not a viable economic model for a feature that has become a commodity on my discussion sites.
I hear ya. I can see why the credit limits might feel restrictive at first.
A few things that might help put this in perspective:
Credits replenish daily, so you get a fresh 15K (Starter) or 30K (Pro) every 24 hours. In practice, it’s pretty unlikely you’d see 20 different users all requesting summaries on the same day, especially since we have caching in place. Once a topic is summarized, subsequent users see the cached version without consuming additional credits. But if you genuinely are seeing that level of daily summary usage, that’s actually a great sign! It means your forum is really active and people are engaged.
If usage really is that high, you might be ready for a higher tier. The Business tier (100k credits) would comfortably support a very active community. High AI feature usage usually signals you’re outgrowing your current plan in other ways too.
Before you consider a tier upgrade, per-user quotas can also help. The LLM quota system lets you distribute usage more sustainably across your users, so a few power users don’t exhaust everything early in the day.
If that still isn’t helpful, third-party LLMs might be a better fit for your use case. You can connect your own LLM provider (OpenAI, Anthropic, Gemini, etc.). You’d be paying for it separately, but it gives you more control and might be more economical for high-volume usage.
We’re trying to offer an out-of-the-box option that works well for most customers without the hassle of API keys or separate billing. But you’re not restricted to using it. It’s meant to be helpful, and if it feels too limiting, switching to a third-party provider is totally valid and we support that fully.
Enabling the hosted LLM
Usage and limits