Add support for self-hosted AI solutions such as Ollama in OpenAI Assistants

Agree x 100. While the OpenAI API service is useful to be able to quickly try out this module and explore AI assistants, and may be a good solution for some projects, many other projects will need self hosted AI or local AI.

My current thinking is that we need middleware that can provide a common API for multiple LLMs. That could be a module that provides integration with existing middleware or a module that provides the API abstraction itself.

Thanks for pointing me to AI Interpolator, I'll have a look at it's approach.

Hi @Robert Castelo,

provide a common API for multiple LLMs

For this I initiated https://www.drupal.org/project/aimodels → but deprecated in favor of having many modules, like these:

• https://www.drupal.org/project/ollama →
• https://www.drupal.org/project/huggingface →

While cost calculation part of aimodels can have a separate module #3390625-2: Debugging / logging feature for prompt engineering →

And building AI assistants feature could be done in Drupal natively:

• https://www.drupal.org/project/aichat →
• https://www.drupal.org/project/aipersona →
• https://www.drupal.org/project/aiwork → (or extended/replaced by AI Interpolator)

Thanks for the advice @mindaugasd

By coincidence when I got your comment I was on a call with @Marcus_Johansson discussing AI Interpolator.

I will definitely be exploring the Huggingface module as an abstraction layer.

The other modules I may not be able to use for the moment until they get at least a beta release as most of what I'm developing will soon go on to a production site.

Since our discussion, there was a new promising module for abstraction created https://www.drupal.org/project/llm_provider →
We had some discussion about its future here: ✨ Collaboration with existing projects Active
Ideally, this would be the only module anyone would need to integrate all LLMs.

I asked @Marcus_Johansson and he is interested to use "LLM provider" for abstraction as well, but I think it depends when LLM provider will release. Please let @seogow know if you are interested in the module.