博查万象

What is Bocha Wanxiang

Bocha Wanxiang is a next-generation search engine based on multimodal hybrid search and semantic reranking technology, providing API services such as Web Search, AI Search, and Semantic Reranker. Bocha Wanxiang is based on multimodal hybrid search and semantic reranking technology, retrieving high-quality content such as news, images, videos, and encyclopedias from nearly 10 billion web pages, supporting natural language queries and 0.15-second ultra-fast responses, with data staying within China and complying with domestic security standards, helping AI Agents, Chatbots, and RAG applications reduce hallucinations and connect to real-time world knowledge.

Main Features of Bocha Wanxiang

• Web Search API (General Search): Supports natural language search across the entire web, returning structured data such as titles, URLs, summaries, website icons, publication times, and images, with up to 50 results per request and response latency as low as 0.15 seconds.
• AI Search API (AI Search): Hybrid search across the entire web with automatic retrieval of vertical-domain structured data, supporting real-time answer generation by large models, follow-up questions, and streaming output, returning multimodal reference sources and summarized answers.
• Semantic Reranker API (Semantic Reranking): Self-developed 80M-parameter Transformer model that semantically reranks candidate results after BM25/RRF initial ranking, supporting recall50@top_n50 ranking, with results better aligned with LLM preferences.
• Agent Search API (Agent Search): Calls a specified AI agent to answer user questions, returning professional in-depth answers and follow-up questions (closed beta ended; vertical-domain needs can switch to AI Search).
• Bocha Hybrid Search Database（BochaDB）: Supports keyword + vector hybrid search, with built-in semantic reranking capability, providing one-stop retrieval infrastructure for RAG applications.
• Balance Query API: Real-time query of account balance and resource consumption.

How to Use Bocha Wanxiang

• Register and Obtain an API-KEY: Visit the Bocha Wanxiang official website https://open.bochaai.com/ to register an account, claim a free call resource package, and obtain an API-KEY for subsequent API authentication.
• Choose an Integration Method: The platform supports four calling methods: official REST API, large model Function Call (compatible with OpenAI / LangChain), AI development platform plugins (Tencent Yuanqi、Coze, DingTalk,Dify), and MCP Server.
• Call the Web Search API (General Search): Via https://api.bocha.cn/v1/web-search for web-wide search, supporting natural language queries, specified time ranges, limited website scopes, and output text summaries.
• Call the AI Search API (AI Search): Perform multimodal hybrid search, automatically retrieve vertical-domain structured data, support real-time answer generation by large models, follow-up questions, and streaming output, and return web pages, images, multimodal reference sources (modality cards), and summarized answers.
• Call the Semantic Reranker API (Semantic Reranking): Use the self-developed bocha-semantic-reranker to perform semantic reranking on existing candidate documents, supporting recall50@top_n50 ranking, outputting relevance scores and ranking results to make results better aligned with LLM preferences.
• Query Account Balance: Via the balance query API, view account resource consumption and remaining quota in real time, facilitating cost management and usage monitoring.

Key Information and Usage Requirements of Bocha Wanxiang

• API Endpoints：https://api.bocha.cn/v1/web-search , etc., in RESTful JSON format.
• Authentication Method：Bearer Token（API-KEY）。
• Data Coverage: Nearly ten billion web pages and ecosystem content sources, covering verticals such as news, images, videos, encyclopedias, flights and hotels, and academic content.
• Compliance and Security: Data sources and search results comply with domestic security regulations, data does not go overseas, meeting the compliance needs of B2B enterprises.
• Response Performance: Web Search supports ultra-fast responses of 0.15 seconds; AI Search supports both streaming and non-streaming output.
• Return Limit: A single search returns up to 50 web page results (count=50).
• Integration Ecosystem: Provides four ways to call: official API, Function Call, AI development platform plugins, and MCP Server.

Core Advantages of Bocha Wanxiang

• Optimized for AI: Search result formats are designed specifically for large models. Semantic Reranker and LLM both use the Transformer architecture, and the reranked results are more favored by LLMs.
• Accurate Hybrid Search: Intent recognition Router drives keyword + vector multi-path recall, with BM25/RRF initial ranking and semantic fine ranking, improving search accuracy across the entire pipeline.
• Lightweight and Efficient: Self-developed 80M parameter semantic ranking model, with performance approaching world-leading 280M/560M parameter models, faster inference speed, and lower cost.
• Compliance Leadership: A rare content-compliant search API in China, data does not go overseas, filling a gap in the domestic market.
• Ultra-Fast Response: Web Search supports 0.15-second response latency, meeting high-concurrency real-time retrieval scenarios.
• Comprehensive Ecosystem Integration: Native support for OpenAI/LangChain Function Call and MCP Server, and provides plugins for intelligent agent platforms such as Tencent Yuanqi, Coze, DingTalk, and Dify.

Comparison of Bocha Wanxiang with Similar Competitors

Application scenarios of Bocha Wanxiang

• AI dialogue enhancement: Connect real-time web search capabilities to Chatbots, reduce large model hallucinations based on high-quality retrieval context, and improve the timeliness and accuracy of responses.
• RAG knowledge Q&A: As an external retrieval module of the RAG system, recall structured knowledge with citation sources from nearly ten billion web pages, supporting enterprise knowledge bases and open-domain question answering.
• AI search engine: Build a conversational search product with summaries, multimodal reference sources, and follow-up questions, supporting streaming output and vertical-domain structured data display.
• Agent development: Through Function Call or platform plugins (Tencent Yuanqi/Coze/DingTalk/Dify), give AI Agents real-time information acquisition and deep reasoning capabilities.
• Enterprise compliance retrieval: Meet the security requirements of domestic B2B enterprises for content compliance and no data going overseas, replacing overseas search APIs for localized AI applications.