AI Search Open Platform is an open and flexible service platform. Beyond atomic capabilities, it lets you connect individual capabilities into pipelines and then combine those pipelines into complete solutions. You can go from a question to task execution with a single call, and you can customize the entire pipeline as needed. Its core capabilities are organized into three layers: 1) Component-based algorithm and model services, which are core capability components such as document parsing, text embedding, multimodal embedding, web search, and LLMs. 2) General-purpose offline data processing pipelines that combine multiple component-based capabilities, such as the RAG pipeline and the multimodal image search pipeline. 3) Out-of-the-box end-to-end solutions, such as Agentic Search. Whether you are an enterprise that wants to deploy end-to-end applications quickly or a developer who needs underlying algorithm services for secondary development, AI Search Open Platform provides the best path.
-
You can activate AI Search Open Platform for free. You are not charged if you do not use it.
-
After a new user activates AI Search Open Platform, the system provides each Alibaba Cloud account with a free quota of 10 service calls. An Alibaba Cloud account is also called a main account. The main account and its RAM users share the free quota. Click Activate Now to try the model services. After the 10 free calls are used up, you are charged based on your actual model service usage.
Capability 1: Component-based AI search services
For developers and ISVs, AI Search Open Platform provides modular, composable AI search capability components that cover the complete pipeline from content parsing to result generation. These components support rapid development for cutting-edge scenarios such as RAG and multimodal search.
Data parsing services
Document parsing service
Parses documents and images within minutes. For documents such as PDF, DOC, HTML, and TXT files, the service can distinguish between multiple layouts and extract logical hierarchical structures, such as titles and paragraphs, from unstructured documents. It also extracts information such as text, tables, images, and code, removes headers and footers, recognizes superscripts and subscripts, and outputs the results in a structured format.
Image parsing service
Provides image content understanding for image data such as architecture diagrams and analytical charts. The service can use multimodal large models to parse and understand image content and recognize text, or use OCR capabilities to recognize and extract text from images. The extracted information can be used in scenarios such as image retrieval and Q&A.
Object detection service
Automatically detects and recognizes one or more main objects in images. This service is suitable for typical computer vision scenarios such as intelligent surveillance, autonomous driving, and image retrieval.
Document chunking service
Provides a general-purpose document chunking service that splits documents based on document semantics, paragraph structure, and specified rules to improve the efficiency of subsequent document processing and retrieval. The output chunk tree can be used to complete context during retrieval.
Speech recognition service
Provides automatic speech recognition (ASR), which quickly converts speech in videos or audio into structured text. This service can be used in scenarios such as meeting minutes, video retrieval, and online customer service.
Video frame extraction service
Provides video frame extraction, which extracts keyframes from videos. Combined with optical character recognition (OCR), image parsing, or multimodal embedding services, it enables in-depth parsing and structured processing of video content.
Video segmentation service
Understands and analyzes videos, extracts keyframes, and splits videos into corresponding segments.
Video summarization service
Understands specified video segments and uses large model capabilities to generate video summaries, titles, and tags.
Embedding services
Text embedding service
-
Text embedding converts text data into dense vector representations. The service supports multiple text embedding models with different languages, input lengths, and output dimensions, and can be used in scenarios such as information retrieval, text classification, and similarity comparison.
-
Sparse text embedding converts text data into sparse vector representations. Sparse vectors require less storage space and are commonly used to represent keywords and term frequency information. They can be combined with dense vectors for hybrid retrieval to improve the final retrieval performance.
-
The embedding fine-tuning service provides embedding model optimization. By training a custom dimensionality reduction model, it helps reduce high-dimensional vectors to lower dimensions without significant loss of retrieval performance, which improves cost-effectiveness.
Multimodal embedding service
Provides multiple multimodal embedding models that accept text, images, and combined text and image input. These models convert data from different modalities into semantic vectors for scenarios such as cross-modal retrieval.
Search and ranking services
Search engine
Provides vector retrieval and text retrieval engines for storing vector and text content, building indexes, and performing online vector and text retrieval. After you activate the engine service, you can use it together with the wide range of API services provided by AI Search Open Platform.
Ranking service
Provides a relevance ranking service for queries and documents. In RAG and search scenarios, the service identifies more relevant content and returns it in ranked order, helping improve the accuracy of retrieval and large language model generation.
Multimodal ranking service
Provides relevance ranking for images. In RAG and multimodal search scenarios, the ranking service finds more relevant content and returns it in order of relevance. Introducing the ranking service effectively improves the accuracy of retrieval and large model generation.
Large model and AI services
Query analysis service
Provides query content analysis. Based on large language models and NLP capabilities, the service performs intent recognition, similar question expansion, and NL2SQL processing on user queries, which effectively improves retrieval and Q&A performance in RAG scenarios.
Large model service
Provides multiple large language model services, including DeepSeek-V4-Pro, Qwen-Plus, and Qwen-Max. The platform also includes the OpenSearch-Qwen-Turbo model, which is built on the qwen-turbo large language model and enhanced through supervised fine-tuning to strengthen retrieval-augmented generation (RAG) capabilities and reduce hallucinations.
Web search service
Provides web search. When a private knowledge base cannot provide an answer, you can extend the search to the web to obtain more information from the Internet, supplement the private knowledge base, and generate richer answers with large language models.
Benefits
-
Rich AI search capabilities: Dedicated AI search models are trained on leading foundation models, and component-based services are built in for the full pipeline of search and RAG scenarios.
-
Flexible invocation: You can call services by using APIs or SDKs, which makes it easy for developers, enterprise customers, and ISV engineers to integrate part or all of the AI search pipeline into their own business workflows.
-
Out-of-the-box: After activation, you can flexibly call all services.
-
Best practices: Built on years of OpenSearch expertise in intelligent search and RAG, the platform includes multiple built-in AI search best practices that help you quickly build search pipelines tailored to your business needs.
Capability 2: Offline data processing and scenario building
By combining multiple component-based capabilities, AI Search Open Platform provides general-purpose offline data processing pipelines, such as the RAG pipeline and the multimodal search pipeline. Enterprise customers can call these services to quickly process data and build scenarios. Examples:
RAG business development
Scenarios:
-
Intelligent customer service
-
Conversational search
-
Knowledge graph enhancement
-
Personalized recommendation
For a development example, see 
Build a Q&A system based on RAG.
Multimodal search business development
Scenarios:
-
Intelligent search for e-commerce and retail
-
Content and news search
-
Gaming industry search
-
Healthcare industry search
-
Financial industry search
For a development example, see Multimodal search business development
Capability 3: Out-of-the-box end-to-end applications
Based on multiple capability components and business scenarios, AI Search Open Platform provides Agentic Search, a complete end-to-end solution. Agentic Search integrates deep retrieval and research capabilities, adds upgraded task execution capabilities, and deeply integrates search, reasoning, planning, and tools. This solution is suitable for business scenarios such as intelligent customer service, industry research, and enterprise productivity improvement.
For a detailed introduction to the product capabilities, see Agentic Search: the AI-driven next-generation enterprise search agent.