Selfhosting my AI chats: more privacy and sovereignty

I have a lot of fun with self-hosting lately. It got much easier with the help of AI. After tackling and improving music self-hosting, it was finally time to find an adequate solution for my AI chats. Previously I mainly used Gemini, then Mistral Vibe (Le Chat). But I’m not happy with having just another subscription, given I also use OpenCode with Cortecs.ai or OpenRouter for my private programming projects. DeepSeek V4.1 Flash and other open-weight models got so good and cheap that I really see no reason to pay ten, twenty, or even more euros each month to just have a casual chat with an LLM without running into limits. Cortecs.ai has the provider Entrim with a token pricing of €0.180 per million for input and €0.539 per million for output. That’s a lot of chatting and hobby programming before reaching the price of some usual subscriptions. That’s why I now have a setup consisting of Open WebUI on my homeserver, Cortecs.ai as the LLM gateway, and DeepSeek V4.1 Flash as the model. In…