Explain the problem as you see it
Currently, for Tana AI features, the only option is to use OpenAI's GPT.
Why is this a problem for you?
I don't think that GPT is the best AI LLM solution and I'm preferring to use other options.
Suggest a solution
Allow a user to select from different AI models, such as Anthropic's Claude 2, and Inflection's Inflection-1 that will be used for the AI features. Potentially even enable greater customizability and let a user choose the specific API calls that are sent with AI commands to allow for a user to use and fine tune their own model if preferred.
14 Comments
The larger context window for Claude is a huge advantage for working in Tana
Do we know if there's a limit, or whether there is an official API planned soon? Afaik, the API is pretty much unofficial, and also doesn't provide endpoints like openAI does
I would also like to see the integration of the Perplexity API.
Has taken on new significance in recent days (March 2024) with the release of Claude 3, which I'd love to use inside Tana!
Any updates on this please
Even just Claude 3 is fine
Claude 3 is great, but I found Gemini 1.5 Pro even better. (I use it via poe.com.)
Would be great if Tana permitted openrouter.ai as an API-interface to access different language models.
Here's an explanation how harpa.ai uses openrouter, for instance: https://harpa.ai/guides/openrouter-connection-in-harpa-ai
Yes, Claude 3 and Gemini 1.5 definetely needed.
+1 on the openrouter.ai API-interface, or something similar. Would love to have a choice in what model to use. OpenAI is kind of iffy imo.
@Alexander Schwarz is right— there's a super-easy way Tana could do this. Make the OpenAI base URL a system setting instead of hard-coded. Boom! Now we can point it to our own router.ai setup and use any model we want.
I like Claude better than ChatGPT for many if not most tasks... and increasingly, I would also prefer to financially support Anthropic over OpenAI.
I wish Tana could make the OpenAI base URL a system setting. I saw a similar feature request 10 months ago but no updates later.
I second the perplexity comment as well !
I would like to see this for Gemini. I have the Pro version already and would love to use it here.
Fully second all of that and would like to add that it would be great to be able to use an offline LLM - configured my PC especially so that it could run local LLMs - even just the ability to mimic openai calls would be great - as afr as I understand we can currently only input an API key that is directly put to the OpenAI servers(?)
Yeah, hoping we get the ability to bring our own OpenRouter key and just use whatever models we want.