Ability to use local model

[anngvu]

Related to #34.This would be the holy grail for this maintainer, but implied in this is that we would need a trained local model, since local models are terrible with the tasks we need unless fine-tuned (from previous experiments). However, fine-tuning takes GPU resources and dedicated developer time, which we currently don't have. But a local model potentially has these advantages:

• Lower latency for users than with web service-based model providers (though tokens/minute can vary a lot depending on local model training and local hardware).
• Avoid third-party model provider downtime. This can happen quite often with ChatGPT.
• Aligns better with local-first ideology and user privacy concerns.

[allaway]

@anngvu IDK if you have seen this (or maybe you introduced me to this...can't recall) but this might be a nice playground to experiment with local LLMs https://lmstudio.ai/