How it works
When you send a message from this page, the app calls:POST /api/assistant/message
- Looks up your current assistant from Supabase.
- Uses the assistant’s
instructionsas a system prompt. - Streams a reply from the selected model (base or fine‑tuned).
- Stores the conversation in a
chats+messagestable pair for history.
- Model label (e.g.
gpt-4ovs a fine‑tuned model name) - Whether a fine‑tuned model is active (depended on the last successful job)
When to use it
Use the Assistant console to:- Sanity‑check the assistant’s behavior before exposing the API
- Compare responses before and after fine‑tuning
- Reproduce problematic prompts so you can improve your specification

