logancyang
commented
9 months ago @taneltm thanks for the suggestion! I've been thinking about ways to use the Assistant API and what you described is a great way. I'm also thinking about changing "QA: Active Note" to "Local RAG" where it should support any model not limited to OpenAI ones, and will work on a collection of notes with local and non-local models. So in the future, we will have 3 modes:
- Conversation: simple conversation without RAG, supports up to the context window and can include multiple notes in the prompt
- OpenAI Assistant: OpenAI's version of RAG, adopts the new stateful Assistant API using threads, only works with OpenAI models
- Local RAG: powered by any LLM + a local vector store, the most customizable experience
How does this sound? Any comment or suggestion is appreciated!
taneltm
michalzmuda
I have several system prompts which I currently need to copy-paste to the copilot settings page. It would be more convenient if I could just switch between the Assistants I've defined in the OpenAI playground.
Maybe the OpenAI Assistants could simply be selected from the chat mode selection, the one which currently has "Conversation" and "QA: Active Note".
The assistant defines the model, so I guess the model selection should display the model which the assistant uses, with the field itself being disabled.
API: https://platform.openai.com/docs/api-reference/assistants