Discourse offers 3 AI deployment paths:
- BYO API key (OpenAI/Anthropic/Google) → fast setup, low switching cost
- Managed AI → no keys, no ops; runs open-weight models on Discourse infra (data stays put)
- Self-hosted models → full sovereignty + fine-tuning + regulated use cases
One-size AI ages badly. Optionality wins.
https://blog.discourse.org/2026/03/three-ways-to-deploy-ai-in-community-platforms-and-why-choice-matters/
@Discourse Honestly, y'all's heavy focus on AI is *SUPER* off putting, but I'm still willing to use Discourse as long as NO AI AT ALL also remains an option. I greatly appreciate that as an admin, all AI functionality is currently contained within a plugin and can be easily disabled, and I wish y'all would promote that as a feature too!! If the ability to disable AI features was ever taken away, Discourse would honestly be completely dead to me. I also wish *users* had greater flexibility in turning off/hiding that crap for themselves when the admins accidentally or intentionally turned it on. Legitimately the only thing I ever want to ask an AI to do is to tell it to fuck off!