渲染错误: Mermaid 渲染失败: Parsing failed: Lexer error on line 2, column 24: unexpected character: ->(<- at offset: 41, skipped 1 characters. Lexer error on line 2, column 41: unexpected character: ->)<- at offset: 58, skipped 2 characters. Lexer error on line 2, column 49: unexpected character: ->控<- at offset: 66, skipped 4 characters. Lexer error on line 3, column 22: unexpected character: ->[<- at offset: 92, skipped 8 characters. Lexer error on line 4, column 21: unexpected character: ->(<- at offset: 138, skipped 1 characters. Lexer error on line 4, column 25: unexpected character: ->精<- at offset: 142, skipped 8 characters. Lexer error on line 4, column 41: unexpected character: ->存<- at offset: 158, skipped 3 characters. Lexer error on line 5, column 21: unexpected character: ->(<- at offset: 197, skipped 1 characters. Lexer error on line 5, column 25: unexpected character: ->语<- at offset: 201, skipped 8 characters. Lexer error on line 5, column 40: unexpected character: ->向<- at offset: 216, skipped 4 characters. Lexer error on line 6, column 21: unexpected character: ->(<- at offset: 256, skipped 1 characters. Lexer error on line 6, column 25: unexpected character: ->知<- at offset: 260, skipped 8 characters. Lexer error on line 6, column 41: unexpected character: ->结<- at offset: 276, skipped 6 characters. Lexer error on line 7, column 22: unexpected character: ->(<- at offset: 319, skipped 6 characters. Lexer error on line 8, column 28: unexpected character: ->(<- at offset: 370, skipped 8 characters. Lexer error on line 9, column 24: unexpected character: ->(<- at offset: 419, skipped 7 characters. Lexer error on line 10, column 25: unexpected character: ->(<- at offset: 468, skipped 6 characters. Lexer error on line 11, column 27: unexpected character: ->(<- at offset: 518, skipped 14 characters. Lexer error on line 13, column 31: unexpected character: ->用<- at offset: 564, skipped 4 characters. Lexer error on line 14, column 30: unexpected character: ->.<- at offset: 598, skipped 1 characters. Lexer error on line 14, column 32: unexpected character: ->精<- at offset: 600, skipped 6 characters. Lexer error on line 15, column 29: unexpected character: ->命<- at offset: 635, skipped 4 characters. Lexer error on line 16, column 30: unexpected character: ->.<- at offset: 669, skipped 1 characters. Lexer error on line 16, column 32: unexpected character: ->未<- at offset: 671, skipped 10 characters. Lexer error on line 17, column 29: unexpected character: ->命<- at offset: 710, skipped 4 characters. Lexer error on line 18, column 30: unexpected character: ->.<- at offset: 744, skipped 1 characters. Lexer error on line 18, column 32: unexpected character: ->未<- at offset: 746, skipped 10 characters. Lexer error on line 19, column 36: unexpected character: ->构<- at offset: 792, skipped 7 characters. Lexer error on line 20, column 38: unexpected character: ->调<- at offset: 837, skipped 5 characters. Lexer error on line 21, column 35: unexpected character: ->按<- at offset: 877, skipped 6 characters. Lexer error on line 22, column 37: unexpected character: ->按<- at offset: 920, skipped 7 characters. Lexer error on line 23, column 32: unexpected character: ->返<- at offset: 959, skipped 6 characters. Lexer error on line 24, column 29: unexpected character: ->写<- at offset: 994, skipped 8 characters. Lexer error on line 25, column 29: unexpected character: ->写<- at offset: 1031, skipped 6 characters. Lexer error on line 26, column 29: unexpected character: ->沉<- at offset: 1066, skipped 7 characters. Lexer error on line 27, column 25: unexpected character: ->返<- at offset: 1098, skipped 4 characters. Parse error on line 2, column 25: Expecting: one of these possible Token sequences: 1. [NEWLINE] 2. [EOF] but found: 'AI' Parse error on line 2, column 28: Expecting token of type ':' but found `Agent`. Parse error on line 2, column 34: Expecting: one of these possible Token sequences: 1. [NEWLINE] 2. [EOF] but found: 'Harness' Parse error on line 2, column 43: Expecting token of type ':' but found `Agent`. Parse error on line 4, column 22: Expecting: one of these possible Token sequences: 1. [NEWLINE] 2. [EOF] but found: 'L' Parse error on line 4, column 33: Expecting token of type ':' but found `R`. Parse error on line 4, column 34: Expecting: one of these possible Token sequences: 1. [--] 2. [-] but found: 'edis' Parse error on line 4, column 39: Expecting: one of these possible Token sequences: 1. [NEWLINE] 2. [EOF] but found: 'KV' Parse error on line 4, column 45: Expecting token of type ':' but found `in`. Parse error on line 5, column 22: Expecting: one of these possible Token sequences: 1. [NEWLINE] 2. [EOF] but found: 'L' Parse error on line 5, column 33: Expecting token of type ':' but found `Milvus`. Parse error on line 5, column 45: Expecting: one of these possible Token sequences: 1. [NEWLINE] 2. [EOF] but found: 'in' Parse error on line 5, column 59: Expecting token of type ':' but found ` `. Parse error on line 6, column 22: Expecting: one of these possible Token sequences: 1. [NEWLINE] 2. [EOF] but found: 'L' Parse error on line 6, column 33: Expecting token of type ':' but found `MongoDB`. Parse error on line 6, column 48: Expecting: one of these possible Token sequences: 1. [NEWLINE] 2. [EOF] but found: 'in' Parse error on line 6, column 62: Expecting token of type ':' but found ` `. Parse error on line 13, column 9: Expecting token of type ':' but found `(User)`. Parse error on line 13, column 20: Expecting token of type 'ARROW_DIRECTION' but found `scheduler`. Parse error on line 13, column 29: Expecting: one of these possible Token sequences: 1. [NEWLINE] 2. [EOF] but found: ':' Parse error on line 14, column 15: Expecting token of type ':' but found `--`. Parse error on line 14, column 19: Expecting token of type 'ARROW_DIRECTION' but found `l1_cache`. Parse error on line 14, column 27: Expecting: one of these possible Token sequences: 1. [NEWLINE] 2. [EOF] but found: ':' Parse error on line 14, column 38: Expecting token of type ':' but found ` `. Parse error on line 15, column 14: Expecting token of type ':' but found `--`. Parse error on line 15, column 18: Expecting token of type 'ARROW_DIRECTION' but found `scheduler`. Parse error on line 15, column 27: Expecting: one of these possible Token sequences: 1. [NEWLINE] 2. [EOF] but found: ':' Parse error on line 16, column 15: Expecting token of type ':' but found `--`. Parse error on line 16, column 19: Expecting token of type 'ARROW_DIRECTION' but found `l2_cache`. Parse error on line 16, column 27: Expecting: one of these possible Token sequences: 1. [NEWLINE] 2. [EOF] but found: ':' Parse error on line 16, column 42: Expecting token of type ':' but found ` `. Parse error on line 17, column 14: Expecting token of type ':' but found `--`. Parse error on line 17, column 18: Expecting token of type 'ARROW_DIRECTION' but found `scheduler`. Parse error on line 17, column 27: Expecting: one of these possible Token sequences: 1. [NEWLINE] 2. [EOF] but found: ':' Parse error on line 18, column 15: Expecting token of type ':' but found `--`. Parse error on line 18, column 19: Expecting token of type 'ARROW_DIRECTION' but found `l3_cache`. Parse error on line 18, column 27: Expecting: one of these possible Token sequences: 1. [NEWLINE] 2. [EOF] but found: ':' Parse error on line 18, column 42: Expecting token of type ':' but found ` `. Parse error on line 19, column 15: Expecting token of type ':' but found `--`. Parse error on line 19, column 19: Expecting token of type 'ARROW_DIRECTION' but found `context_manager`. Parse error on line 19, column 34: Expecting: one of these possible Token sequences: 1. [NEWLINE] 2. [EOF] but found: ':' Parse error on line 20, column 21: Expecting token of type ':' but found `--`. Parse error on line 20, column 25: Expecting token of type 'ARROW_DIRECTION' but found `llm_gateway`. Parse error on line 20, column 36: Expecting: one of these possible Token sequences: 1. [NEWLINE] 2. [EOF] but found: ':' Parse error on line 21, column 17: Expecting token of type ':' but found `--`. Parse error on line 21, column 21: Expecting token of type 'ARROW_DIRECTION' but found `tool_gateway`. Parse error on line 21, column 33: Expecting: one of these possible Token sequences: 1. [NEWLINE] 2. [EOF] but found: ':' Parse error on line 22, column 17: Expecting token of type ':' but found `--`. Parse error on line 22, column 21: Expecting token of type 'ARROW_DIRECTION' but found `knowledge_base`. Parse error on line 22, column 35: Expecting: one of these possible Token sequences: 1. [NEWLINE] 2. [EOF] but found: ':' Parse error on line 23, column 17: Expecting token of type ':' but found `--`. Parse error on line 23, column 21: Expecting token of type 'ARROW_DIRECTION' but found `scheduler`. Parse error on line 23, column 30: Expecting: one of these possible Token sequences: 1. [NEWLINE] 2. [EOF] but found: ':' Parse error on line 24, column 15: Expecting token of type ':' but found `--`. Parse error on line 24, column 19: Expecting token of type 'ARROW_DIRECTION' but found `l1_cache`. Parse error on line 24, column 27: Expecting: one of these possible Token sequences: 1. [NEWLINE] 2. [EOF] but found: ':' Parse error on line 25, column 15: Expecting token of type ':' but found `--`. Parse error on line 25, column 19: Expecting token of type 'ARROW_DIRECTION' but found `l2_cache`. Parse error on line 25, column 27: Expecting: one of these possible Token sequences: 1. [NEWLINE] 2. [EOF] but found: ':' Parse error on line 26, column 15: Expecting token of type ':' but found `--`. Parse error on line 26, column 19: Expecting token of type 'ARROW_DIRECTION' but found `l3_cache`. Parse error on line 26, column 27: Expecting: one of these possible Token sequences: 1. [NEWLINE] 2. [EOF] but found: ':' Parse error on line 27, column 15: Expecting token of type ':' but found `--`. Parse error on line 27, column 19: Expecting token of type 'ARROW_DIRECTION' but found `user`. Parse error on line 27, column 23: Expecting: one of these possible Token sequences: 1. [NEWLINE] 2. [EOF] but found: ':'

from llmlingua import PromptCompressor
import jieba
from typing import List, Dict
import hashlib

# 初始化LLMLingua压缩器，中文场景用bert-base模型，平衡速度和效果
compressor = PromptCompressor(
    model_name_or_path="microsoft/llmlingua-2-bert-base-chinese",
    device_map="cpu"
)

def extract_context_fingerprint(context: List[Dict]) -> str:
    """
    提取上下文核心指纹：提取用户ID、商品ID、核心意图等关键变量，生成哈希
    """
    core_vars = []
    for msg in context:
        if msg.get("role") == "system" and "user_id" in msg["content"]:
            core_vars.append(msg["content"].split("user_id:")[1].split("\n")[0].strip())
        if "product_id" in msg["content"]:
            core_vars.append(msg["content"].split("product_id:")[1].split("\n")[0].strip())
    core_str = "|".join(core_vars)
    return hashlib.md5(core_str.encode("utf-8")).hexdigest()

def filter_redundant_content(context: List[Dict]) -> List[Dict]:
    """
    过滤上下文冗余内容：重复语句、无意义符号、工具返回的无效调试信息
    """
    seen_contents = set()
    filtered_context = []
    for msg in context:
        content = msg["content"].strip()
        if not content:
            continue
        # 去重
        content_hash = hashlib.md5(content.encode("utf-8")).hexdigest()
        if content_hash in seen_contents:
            continue
        seen_contents.add(content_hash)
        # 过滤工具返回的调试信息
        if msg.get("role") == "tool" and "debug_info" in content:
            content = content.split("debug_info")[0].strip()
            msg["content"] = content
        filtered_context.append(msg)
    return filtered_context

def compress_context(context: List[Dict], target_token: int = 1024) -> List[Dict]:
    """
    上下文压缩入口函数，返回压缩后的上下文
    """
    filtered_context = filter_redundant_content(context)
    # 转成字符串格式
    context_str = "\n".join([f"{msg['role']}: {msg['content']}" for msg in filtered_context])
    # 压缩，保留关键分隔符
    compressed_result = compressor.compress_prompt(
        context_str,
        target_token=target_token,
        rate=0.3,
        force_tokens=["\n", "user:", "assistant:", "system:", "tool:"]
    )
    # 转回上下文格式
    compressed_context = []
    for line in compressed_result["compressed_prompt"].split("\n"):
        if line.startswith("system:"):
            compressed_context.append({"role": "system", "content": line.split("system:")[1].strip()})
        elif line.startswith("user:"):
            compressed_context.append({"role": "user", "content": line.split("user:")[1].strip()})
        elif line.startswith("assistant:"):
            compressed_context.append({"role": "assistant", "content": line.split("assistant:")[1].strip()})
        elif line.startswith("tool:"):
            compressed_context.append({"role": "tool", "content": line.split("tool:")[1].strip()})
    return compressed_context

from sentence_transformers import SentenceTransformer
import numpy as np
from pymilvus import MilvusClient

# 初始化embedding模型
embedding_model = SentenceTransformer("BAAI/bge-m3")
# 初始化Milvus客户端
milvus_client = MilvusClient(uri="http://localhost:19530")
COLLECTION_NAME = "agent_semantic_cache"

def calc_cosine_similarity(vec1: np.array, vec2: np.array) -> float:
    """计算两个向量的余弦相似度"""
    return np.dot(vec1, vec2) / (np.linalg.norm(vec1) * np.linalg.norm(vec2))

def search_semantic_cache(query: str, context_fingerprint: str, threshold: float = 0.92) -> str:
    """
    语义缓存查询，返回匹配的结果，没有匹配返回None
    """
    # 生成向量
    query_vec = embedding_model.encode(f"{query}|{context_fingerprint}", normalize_embeddings=True)
    # 检索向量库
    search_result = milvus_client.search(
        collection_name=COLLECTION_NAME,
        data=[query_vec],
        limit=1,
        output_fields=["response"],
        search_params={"metric_type": "COSINE", "params": {"nprobe": 10}}
    )
    if not search_result or len(search_result[0]) == 0:
        return None
    top_result = search_result[0][0]
    if top_result["distance"] >= threshold:
        return top_result["entity"]["response"]
    return None

def write_semantic_cache(query: str, context_fingerprint: str, response: str, ttl: int = 86400*7):
    """写入语义缓存"""
    query_vec = embedding_model.encode(f"{query}|{context_fingerprint}", normalize_embeddings=True)
    milvus_client.insert(
        collection_name=COLLECTION_NAME,
        data=[{
            "vector": query_vec,
            "query": query,
            "context_fingerprint": context_fingerprint,
            "response": response,
            "expire_time": int(time.time()) + ttl
        }]
    )