For now, the most powerful LLM models are privative and run on the cloud. There's a way to make using them less cringe, by at least strapping out personal information on the fly. And of course hosting your own UI with open-webui.

LiteLLM (a self hosted LLM API aggregator) has support for Presidio, a plugin to detect and mask private information. So that you can use LLMs for coding and debugging without leaking IPs, URLs, Bitcoin addresses...

Check the docs: https://docs.litellm.ai/docs/proxy/guardrails/pii_masking_v2

BTW: Google offers Gemini 2.5 flash via API for free.