xAI launches Grok Collections API to upload and search the datasets

xAI has launched the Grok Collections API, a new RAG (Retrieval-Augmented Generation) system that allows developers to upload and search through the datasets. The datasets extend from PDFs and Excel sheets to codebases for knowledge and searches.

The new API will enable developers to create RAG applications with ease of management of the indexing and retrieval infrastructure. For the first week, xAI is offering this API for free and a retrieval priced at a flat rate of $2.50 per 1,000 searches.

xAI has revealed some of the advantages of the Collection API. In terms of indexing, the API uses OCR and layout-aware parsing to extract text while maintaining structure such as the layout of a PDF, hierarchy of an Excel table, or code syntax.

The API manages file uploads, updates, and download actions. The system reindexes files with new changes to keep the collection up to date. The company has shared a broad list of collection types, which you can check here.

For retrieval, Grok 4.1 Fast uses semantic search to relate the meaning and intent behind a query. It also looks for precise term matching for a keyword search. Last but not least, the retrieval also includes a hybrid search system, which provides high accuracy through a combination of keyword and semantic search. The AI company is also supporting a dedicated reranker model and reciprocal rank fusion to improve hybrid searches.

In real-world RAG tasks benchmarks, the xAI Collections API showed big performance gains for Finance and Coding. The tests include Tabular and numerical questions, code understanding, and large file systems. In the meantime, it comes second to Google Gemini Pro 3 in legal RAG tasks.

xAI has revealed that it doesn’t store Collections’ user data for model training without user consent. Check more details about the Grok Collections API on the xAI website.