Setting up through venice.ai for privacy

[AO12612]

TBH I haven't been able to install correctly yet, but I wanted to suggest supporting venice.ai as a first tier option for models for privacy reasons.
They offer both custom hosted uncensored models (fully private) and anonymized access to the main labs models, including opus 4.5.
long term memory is a fundamental feature, but privacy becomes an ever bigger issue the more meaningful the history becomes.

Would be happy to provide keys to support development if needed.
let me know and thank you for your work

[taylorsatula]

Does it not use an OpenAI-compatible endpoint?

[AO12612]

yes, totally.
So users that are aware can make it work.

there might be a few extra features in the api to fine control models (ie turning reasoning on or off by appending a tag to the model name). But those are just nice-to-haves and motivated users can certainly figure it out on their own.

What I was suggesting was to present the privacy option.
Long term memory does profile a user very very deeply.
I believe it is important to present a maximally privacy respecting option in the spirit of open source, until local models are powerful enough to compete wth centralized models for individual's use cases. We will get there, but it will be many years.

https://www.reddit.com/r/technology/comments/1pxealt/openais_chatgpt_ads_will_allegedly_prioritize/
to be honest, it's not even about the enshittification. It's about the information asymmetry and how once data is leaked it never goes back. By the law of unintended consequences I think it would be important to at least try to secure sensitive information.

[AO12612]

btw, thanks for the impressive work.
also you fixed the stylesheet rendering so fast!

[taylorsatula]

If you change the system_llm and account_tier endpoints in the SQL it'll automatically and seamlessly work with OpenAI compatible providers. All translation is done at the @llm_provider.py layer. Perhaps that will help with setting it. Review deploy/config.py for additional insights into configuring it. When a fresh-install users 'offline mode' that is what its doing except it points it to the ollama localhost:11314 endpoint and sets a qwen3:1.7b model.

Also, you're so welcome. I'm glad you're liking it. My logic is that if someone cares enough to send a bug report I care enough to fix it as soon as possible. I appreciate your catch.