Web Analytics

proxyless-llm-websearch

⭐ 122 stars Indonesian by itshyao

🌐 Bahasa

🧠 Mesin Pencari Web LLM Tanpa Proxy

Sebuah alat pencarian web LLM multi-mesin pencari tanpa proxy, mendukung parsing konten URL dan perayapan web, menggabungkan LangGraph dan LangGraph-MCP untuk membangun rantai agen modular. Dirancang khusus untuk skenario penggunaan penarikan pengetahuan eksternal oleh model bahasa besar, mendukung pengambilan dan parsing web dengan Playwright + Crawl4AI, mendukung asinkron, pemrosesan paralel, pemotongan konten, dan filtrasi serta penyusunan ulang.

🚀 Log Pembaruan

✨ Daftar Fitur

workflow

framework

⚡ Mulai Cepat

Kloning Repository

git clone https://github.com/itshyao/proxyless-llm-websearch.git
cd proxyless-llm-websearch

Instalasi Dependensi

pip install -r requirements.txt
python -m playwright install

Konfigurasi Variabel Lingkungan

# 百炼llm
OPENAI_BASE_URL=https://dashscope.aliyuncs.com/compatible-mode/v1
OPENAI_API_KEY=sk-xxx
MODEL_NAME=qwen-plus-latest
百炼embedding

EMBEDDING_BASE_URL=https://dashscope.aliyuncs.com/compatible-mode/v1
EMBEDDING_API_KEY=sk-xxx
EMBEDDING_MODEL_NAME=text-embedding-v4
百炼reranker

RERANK_BASE_URL=https://dashscope.aliyuncs.com/compatible-mode/v1
RERANK_API_KEY=sk-xxx
RERANK_MODEL=gte-rerank-v2

Langgraph-Agent

#### DEMO

python agent/demo.py

#### API SERVE

python agent/api_serve.py

import requests
url = "http://localhost:8800/search"
data = {
  "question": "广州今日天气",
  "engine": "bing",
  "split": {
    "chunk_size": 512,
    "chunk_overlap": 128
  },
  "rerank": {
    "top_k": 5
  }
}
try:
    response = requests.post(
        url,
        json=data
    )
    if response.status_code == 200:
        print("✅ 请求成功！")
        print("响应内容：", response.json())
    else:
        print(f"❌ 请求失败，状态码：{response.status_code}")
        print("错误信息：", response.text)
except requests.exceptions.RequestException as e:
    print(f"⚠️ 请求异常：{str(e)}")

#### Gradio

python agent/gradio_demo.py

gradio

gradio

#### docker

docker-compose -f docker-compose-ag.yml up -d --build

Langgrph-MCP

#### Menyalakan layanan MCP

python mcp/websearch.py

#### DEMO

python mcp/demo.py

#### API SERVE

python mcp/api_serve.py

import requests
url = "http://localhost:8800/search"
data = {
  "question": "广州今日天气"
}
try:
    response = requests.post(
        url,
        json=data
    )
    if response.status_code == 200:
        print("✅ 请求成功！")
        print("响应内容：", response.json())
    else:
        print(f"❌ 请求失败，状态码：{response.status_code}")
        print("错误信息：", response.text)
except requests.exceptions.RequestException as e:
    print(f"⚠️ 请求异常：{str(e)}")

#### docker

docker-compose -f docker-compose-mcp.yml up -d --build

Modul Kustom

#### Pembagian Blok Kustom

from typing import Optional, List
class YourSplitter:
    def __init__(self, text: str, chunk_size: int = 512, chunk_overlap: int = 128):
        self.text = text
        self.chunk_size = chunk_size
        self.chunk_overlap = chunk_overlap
    def split_text(self, text: Optional[str] = None) -> List:
        # TODO: implement splitting logic
        return ["your chunk"]

#### Penataan Ulang Kustom

from typing import List, Union, Tuple
class YourReranker:
    async def get_reranked_documents(
        self,
        query: Union[str, List[str]],
        documents: List[str],
    ) -> Union[
        Tuple[List[str]],
        Tuple[List[int]],
    ]:
        return ["your chunk"], ["chunk index"]

🔍 Perbandingan dengan Pengujian Pencarian Jaringan Online

Kami membandingkan proyek ini dengan beberapa API daring utama untuk mengevaluasi performanya dalam menjawab pertanyaan kompleks.

🔥 Dataset

🧑‍🏫 Hasil Perbandingan

| Mesin Pencari/Sistem | ✅ Benar | ❌ Salah | ⚠️ Sebagian Benar | | -------------------- | -------- | -------- | ---------------- | | Volcano Ark | 5.00% | 72.21% | 22.79% | | Bailian | 9.85% | 62.79% | 27.35% | | Our | 19.85% | 47.94% | 32.06% |

🙏 Ucapan Terima Kasih

Sebagian fungsi dari proyek ini didukung dan terinspirasi oleh proyek open source berikut, dengan ini kami mengucapkan terima kasih:

--- Tranlated By Open Ai Tx | Last indexed: 2025-09-08 ---

Original README: View on GitHub

Translated by OpenAiTx | Last updated: 2025-09-16