# 用自己的文件做一个本地知识库：切块 → 变成向量 → 找最相关的几块 → 交给模型回答。
# 只用 Python 自带的库，模型都跑在本机的 Ollama 上。用法：python3 kb.py "你的问题"
import json, pathlib, sys, urllib.request

OLLAMA = "http://localhost:11434"
EMBED_MODEL = "embeddinggemma"
CHAT_MODEL = "deepseek-r1:8b"

def post(path, payload):
    req = urllib.request.Request(OLLAMA + path, data=json.dumps(payload).encode(),
                                 headers={"Content-Type": "application/json"})
    with urllib.request.urlopen(req, timeout=300) as r:
        return json.load(r)

# 1. 切块：每个文件按行切开，一行就是一块
chunks = []
for f in sorted(pathlib.Path("docs").glob("*.txt")):
    for line in f.read_text(encoding="utf-8").splitlines():
        if line.strip():
            chunks.append({"source": f.name, "text": line.strip()})

# 2. 变成向量：一次把所有块交给嵌入模型
vectors = post("/api/embed", {"model": EMBED_MODEL, "input": [c["text"] for c in chunks]})["embeddings"]

# 3. 找最相关的块：问题也变成向量，和每一块比相似度（向量已经是单位长度，点积就是余弦相似度）
question = sys.argv[1]
q = post("/api/embed", {"model": EMBED_MODEL, "input": question})["embeddings"][0]
scored = sorted(((sum(a * b for a, b in zip(q, v)), c) for v, c in zip(vectors, chunks)),
                key=lambda x: x[0], reverse=True)[:3]
for score, c in scored:
    print(f"[{score:.2f}] {c['source']}：{c['text']}")

# 4. 交给模型回答：把找到的内容放进提示词，要求只根据它们回答
context = "\n".join(f"- {c['text']}" for _, c in scored)
prompt = f"只根据下面的资料回答问题。资料里没有的，就说“资料里没有写”。\n\n资料：\n{context}\n\n问题：{question}"
answer = post("/api/chat", {"model": CHAT_MODEL, "stream": False, "think": False,
                            "messages": [{"role": "user", "content": prompt}]})
print("\n回答：" + answer["message"]["content"].strip())
