orchestrate.py

import ollama
import os
from supabase import create_client
from retrieve import embed_query, match_chunks
from dotenv import load_dotenv


def build_prompt(usr_input, chunks):
    chunks_text = '\n\n'.join(chunks) # 两个换行=空一行，更好区分

    # prompots 之间故意想空一行让 prompt 更清晰( 但浪费token？)
    prompt = f"""Your role: Helpful Assistant

Reference knowledge: {chunks_text} 

User input: {usr_input}

How to respond: based on your knowledge plus the reference knowledge"""
    return prompt

def generate_response(model, prompt):
    # role: user 说的话，system 系统指令， assistant LLM 之前说的话，用于多轮对话
    response = ollama.chat(model = model, messages = [{'role':'user', 'content':prompt}])
    return response['message']['content']

#  ollama 返回对象，大致长这样“
# {
#     'model': 'qwen3:4b',
#     'created_at': '2025-...',
#     'message': {
#         'role': 'assistant',
#         'content': '这里才是LLM真正的回复文本'
#     },
#     'done': True,
#     ...
# }

def main():
    load_dotenv()
    
    SUPABASE_URL = os.getenv('SUPABASE_URL')
    SUPABASE_KEY = os.getenv('SUPABASE_KEY')
    supabase = create_client(SUPABASE_URL, SUPABASE_KEY)
    usr_query = 'test_query'
    model = 'qwen3:4b'
    usr_embed = embed_query(usr_query)
    result = match_chunks(supabase, usr_embed)
    chunks = [row['content'] for row in result.data]
    prompt = build_prompt(usr_query, chunks)
    reply = generate_response(model, prompt)

    return reply

if __name__ == '__main__':
    print(main())