|
| 1 | +"""BochaAI Search API retriever for tree text memory.""" |
| 2 | + |
| 3 | +from concurrent.futures import ThreadPoolExecutor, as_completed |
| 4 | +from datetime import datetime |
| 5 | + |
| 6 | +import requests |
| 7 | + |
| 8 | +from memos.embedders.factory import OllamaEmbedder |
| 9 | +from memos.log import get_logger |
| 10 | +from memos.mem_reader.base import BaseMemReader |
| 11 | +from memos.memories.textual.item import TextualMemoryItem |
| 12 | + |
| 13 | + |
| 14 | +logger = get_logger(__name__) |
| 15 | + |
| 16 | + |
| 17 | +class BochaAISearchAPI: |
| 18 | + """BochaAI Search API Client""" |
| 19 | + |
| 20 | + def __init__(self, api_key: str, max_results: int = 20): |
| 21 | + """ |
| 22 | + Initialize BochaAI Search API client. |
| 23 | +
|
| 24 | + Args: |
| 25 | + api_key: BochaAI API key |
| 26 | + max_results: Maximum number of search results to retrieve |
| 27 | + """ |
| 28 | + self.api_key = api_key |
| 29 | + self.max_results = max_results |
| 30 | + |
| 31 | + self.web_url = "https://api.bochaai.com/v1/web-search" |
| 32 | + self.ai_url = "https://api.bochaai.com/v1/ai-search" |
| 33 | + |
| 34 | + self.headers = { |
| 35 | + "Authorization": f"Bearer {api_key}", |
| 36 | + "Content-Type": "application/json", |
| 37 | + } |
| 38 | + |
| 39 | + def search_web(self, query: str, summary: bool = True, freshness="noLimit") -> list[dict]: |
| 40 | + """ |
| 41 | + Perform a Web Search (equivalent to the first curl). |
| 42 | +
|
| 43 | + Args: |
| 44 | + query: Search query string |
| 45 | + summary: Whether to include summary in the results |
| 46 | + freshness: Freshness filter (e.g. 'noLimit', 'day', 'week') |
| 47 | +
|
| 48 | + Returns: |
| 49 | + A list of search result dicts |
| 50 | + """ |
| 51 | + body = { |
| 52 | + "query": query, |
| 53 | + "summary": summary, |
| 54 | + "freshness": freshness, |
| 55 | + "count": self.max_results, |
| 56 | + } |
| 57 | + return self._post(self.web_url, body) |
| 58 | + |
| 59 | + def search_ai( |
| 60 | + self, query: str, answer: bool = False, stream: bool = False, freshness="noLimit" |
| 61 | + ) -> list[dict]: |
| 62 | + """ |
| 63 | + Perform an AI Search (equivalent to the second curl). |
| 64 | +
|
| 65 | + Args: |
| 66 | + query: Search query string |
| 67 | + answer: Whether BochaAI should generate an answer |
| 68 | + stream: Whether to use streaming response |
| 69 | + freshness: Freshness filter (e.g. 'noLimit', 'day', 'week') |
| 70 | +
|
| 71 | + Returns: |
| 72 | + A list of search result dicts |
| 73 | + """ |
| 74 | + body = { |
| 75 | + "query": query, |
| 76 | + "freshness": freshness, |
| 77 | + "count": self.max_results, |
| 78 | + "answer": answer, |
| 79 | + "stream": stream, |
| 80 | + } |
| 81 | + return self._post(self.ai_url, body) |
| 82 | + |
| 83 | + def _post(self, url: str, body: dict) -> list[dict]: |
| 84 | + """Helper method to send POST request and return JSON results.""" |
| 85 | + try: |
| 86 | + resp = requests.post(url, headers=self.headers, json=body) |
| 87 | + resp.raise_for_status() |
| 88 | + data = resp.json() |
| 89 | + return data.get("results", []) |
| 90 | + except Exception: |
| 91 | + import traceback |
| 92 | + |
| 93 | + logger.error(f"BochaAI search error: {traceback.format_exc()}") |
| 94 | + return [] |
| 95 | + |
| 96 | + |
| 97 | +class BochaAISearchRetriever: |
| 98 | + """BochaAI retriever that converts search results into TextualMemoryItem objects""" |
| 99 | + |
| 100 | + def __init__( |
| 101 | + self, api_key: str, embedder: OllamaEmbedder, reader: BaseMemReader, max_results: int = 20 |
| 102 | + ): |
| 103 | + """ |
| 104 | + Initialize BochaAI Search retriever. |
| 105 | +
|
| 106 | + Args: |
| 107 | + api_key: BochaAI API key |
| 108 | + embedder: Embedder instance for generating embeddings |
| 109 | + reader: MemReader instance for processing internet content |
| 110 | + max_results: Maximum number of search results to retrieve |
| 111 | + """ |
| 112 | + self.bocha_api = BochaAISearchAPI(api_key, max_results=max_results) |
| 113 | + self.embedder = embedder |
| 114 | + self.reader = reader |
| 115 | + |
| 116 | + def retrieve_from_web( |
| 117 | + self, query: str, top_k: int = 10, parsed_goal=None, info=None |
| 118 | + ) -> list[TextualMemoryItem]: |
| 119 | + """Retrieve information using BochaAI Web Search.""" |
| 120 | + search_results = self.bocha_api.search_web(query) |
| 121 | + return self._convert_to_mem_items(search_results, query, parsed_goal, info) |
| 122 | + |
| 123 | + def retrieve_from_ai( |
| 124 | + self, query: str, top_k: int = 10, parsed_goal=None, info=None |
| 125 | + ) -> list[TextualMemoryItem]: |
| 126 | + """Retrieve information using BochaAI AI Search.""" |
| 127 | + search_results = self.bocha_api.search_ai(query) |
| 128 | + return self._convert_to_mem_items(search_results, query, parsed_goal, info) |
| 129 | + |
| 130 | + def _convert_to_mem_items( |
| 131 | + self, search_results: list[dict], query: str, parsed_goal=None, info=None |
| 132 | + ): |
| 133 | + """Convert API search results into TextualMemoryItem objects.""" |
| 134 | + memory_items = [] |
| 135 | + if not info: |
| 136 | + info = {"user_id": "", "session_id": ""} |
| 137 | + |
| 138 | + with ThreadPoolExecutor(max_workers=8) as executor: |
| 139 | + futures = [ |
| 140 | + executor.submit(self._process_result, r, query, parsed_goal, info) |
| 141 | + for r in search_results |
| 142 | + ] |
| 143 | + for future in as_completed(futures): |
| 144 | + try: |
| 145 | + memory_items.extend(future.result()) |
| 146 | + except Exception as e: |
| 147 | + logger.error(f"Error processing BochaAI search result: {e}") |
| 148 | + |
| 149 | + # Deduplicate items by memory text |
| 150 | + unique_memory_items = {item.memory: item for item in memory_items} |
| 151 | + return list(unique_memory_items.values()) |
| 152 | + |
| 153 | + def _process_result( |
| 154 | + self, result: dict, query: str, parsed_goal: str, info: None |
| 155 | + ) -> list[TextualMemoryItem]: |
| 156 | + """Process a single result into one or more TextualMemoryItems.""" |
| 157 | + title = result.get("title", "") |
| 158 | + content = result.get("content", "") |
| 159 | + summary = result.get("summary", "") |
| 160 | + url = result.get("url", "") |
| 161 | + publish_time = datetime.now().strftime( |
| 162 | + "%Y-%m-%d" |
| 163 | + ) # Optional: can map to API field if exists |
| 164 | + |
| 165 | + # Use reader to split and process the content into chunks |
| 166 | + read_items = self.reader.get_memory([content], type="doc", info=info) |
| 167 | + |
| 168 | + memory_items = [] |
| 169 | + for read_item_i in read_items[0]: |
| 170 | + read_item_i.memory = ( |
| 171 | + f"Title: {title}\nNewsTime: {publish_time}\nSummary: {summary}\n" |
| 172 | + f"Content: {read_item_i.memory}" |
| 173 | + ) |
| 174 | + read_item_i.metadata.source = "web" |
| 175 | + read_item_i.metadata.memory_type = "OuterMemory" |
| 176 | + read_item_i.metadata.sources = [url] if url else [] |
| 177 | + read_item_i.metadata.visibility = "public" |
| 178 | + memory_items.append(read_item_i) |
| 179 | + return memory_items |
0 commit comments