One thing I learned is that you really want to be paying for your AI and not using the free versions, because then you can expect to get a professional service that does not use your inputs to train its LLM.
With the Discourse AI I use I depend on the Anthropic API which is not available at no cost anyway.
I created my anthropic account and wired it up with Discourse on February 25, 2026. The cost is dependent on usage, so it can be unpredictable. So far the cost has been manageable at about $60 total for about a month of use. I’ll use this topic to keep an eye on those costs over time.
Things I have done with Claude:
translated my personal forum into German
asked it what it can and can’t do, and what it does with my data
asked it questions about topics discussed on the forum
I took another look at this today because I got another email from anthropic with an invoice of $10 to recharge my tokens, and want to be sure I am not squandering money.
The good news is that so far I have still only spent about $61 on AI since 24 February, according to the Anthropic costs dashboard which looks like this:
You can see from the chart that I spent alot early on. That most likely is when I had all the back history of my personal site going back several years translated into German and also did some early experimentation to find out what it can do and what it can glean from content on my forum.
There was a spike also on Tuesday this week. On that day I used Claude to help me brainstorm what setting up a solopreneurship might look like, and also asked it to help me troubleshoot some technical issues I was grappling with on my self-hosted sites on Hetzner. In both of those areas I got some decent guidance, which I will talk about more elsewhere.
Interestingly, I am able to get much better and more granular stats in one place directly on the Discourse AI usage page. It lets me also track stats across all LLMs I might be using, and gives me a breakdown by feature. The cost is also almost exactly on target! It is able to calculate it because I can specify the costs of various tokens directly in the Discourse AI settings.
So based on this information for today I know I want to turn off the automatic translation, which was just an experiment (everyone in my family, including the Germans, can read English!) and and it was the bulk of the cost. I will turn it on here on this site however as a more public demonstration of how it can work.
Going forward, for the amount of help Claude Sonnet 4.6 is giving me the cost is not bothering me.
I do want to learn about other LLMs and would like to switch to them if they are more ethical and can offer more privacy and safety. I am really appreciating how Discourse AI lets me experiment with different LLMs will still keeping and owning the product of my interactions with them with very little fuss.
