现在做个人知识库基本上是0成本，只需要在google studio上申请一个API Key,几行代码就可以实现完整的问答系统。如果MD文件比较多时，注意调整调用API写入数据库的频率，不控制 十几次就报429错误，加上sleep(1)可解。

embed.py

import google.generativeai as genai
import time
import chunk
import chromadb

EMBEDDING_MODEL = "gemini-embedding-exp-03-07"
LLM_MODEL = "gemini-2.5-flash-preview-05-20"

# 配置 API KEY
API_KEY = "xxxxxxxx-xx-xxxx"
genai.configure(api_key=API_KEY)

chromadb_client = chromadb.PersistentClient("./chroma_gemini.db")
chromadb_collection = chromadb_client.get_or_create_collection("linghuchong_gemini")

def embed(text: str, store: bool) -> list[float]:
    response = genai.embed_content(
        model=EMBEDDING_MODEL,
        content=text,
        task_type="RETRIEVAL_DOCUMENT" if store else "RETRIEVAL_QUERY"
    )
    assert response["embedding"]
    return response["embedding"]

def create_db():
    all_md_files = chunk.get_all_md_files()[:3]  # 只取前3个md文件
    all_chunks = []
    for file_path in all_md_files:
        all_chunks.extend(chunk.get_chunks_from_file(file_path))
        time.sleep(1)  # 每处理一个md文件 sleep 1 秒，避免429
    for idx, c in enumerate(all_chunks):
        print(f"Process: [{c['file']}] {c['content'][:60]}...")
        embedding = embed(c["content"], store=True)
        chromadb_collection.upsert(
            ids=[str(idx)],
            documents=[c["content"]],
            embeddings=[embedding],
            metadatas=[{"file": c["file"]}]
        )
    print("DB created.")

def query_db(question: str) -> list[dict]:
    question_embedding = embed(question, store=False)
    result = chromadb_collection.query(
        query_embeddings=[question_embedding],
        n_results=5
    )
    assert result["documents"]
    docs = result["documents"][0]
    metas = result.get("metadatas", [[]])[0]
    return [{"content": doc, "meta": meta} for doc, meta in zip(docs, metas)]

if __name__ == '__main__':
    import sys
    if len(sys.argv) > 1 and sys.argv[1] == "create":
        create_db()
    else:
        while True:
            question = input("请输入你的问题（输入 exit 退出）：")
            if question.strip().lower() in ("exit", "quit"):
                print("再见！")
                break
            results = query_db(question)
            prompt = "请根据上下文回答用户问题：\n"
            prompt += f"问题：{question}\n"
            prompt += "上下文：\n"
            for r in results:
                prompt += f"[{(r['meta'] or {}).get('file', '')}]\n{r['content']}\n-------------\n"
            model = genai.GenerativeModel(LLM_MODEL)
            response = model.generate_content(prompt)
            print(response.text) 

chunk.py ,如果是复杂的项目可以用 专业分词分词工具 langchain ，做chunk.

 import os
from typing import List, Dict

def get_all_md_files(notes_dir: str = "data/notes") -> List[str]:
    return [
        os.path.join(notes_dir, f)
        for f in os.listdir(notes_dir)
        if f.endswith('.md')
    ]

def read_data(file_path: str) -> str:
    with open(file_path, "r", encoding="utf-8") as f:
        return f.read()

def get_chunks_from_file(file_path: str) -> List[Dict]:
    content = read_data(file_path)
    chunks = content.split('\n\n')
    result = []
    header = ""
    file_name = os.path.basename(file_path)
    for c in chunks:
        if c.strip().startswith("#"):
            header += f"{c.strip()}\n"
        else:
            if c.strip():
                result.append({
                    "content": f"{header}{c.strip()}",
                    "file": file_name
                })
            header = ""
    return result

def get_all_chunks(notes_dir: str = "data/notes") -> List[Dict]:
    all_chunks = []
    for file_path in get_all_md_files(notes_dir):
        all_chunks.extend(get_chunks_from_file(file_path))
    return all_chunks

if __name__ == '__main__':
    for chunk in get_all_chunks():
        print(f"[{chunk['file']}]\n{chunk['content']}\n--------------")