Hologres is a real-time data warehouse with built-in AI capabilities that span the full workflow from data mining and parsing to inference and search. It handles unstructured, structured, and semi-structured data in a single system—no separate AI infrastructure required.
Where do you fit?
Run AI directly in SQL — Call large language models (LLMs) and embedding models from SQL using AI Functions. No separate service to deploy, no data to export.
Build AI-powered search — Combine vector search and full-text search to power similarity search, retrieval-augmented generation (RAG), and enterprise search applications.
Connect AI agents to your data — Use the Hologres MCP Server to let AI agents query your data and metadata through a standard interface.
AI models and AI functions
Hologres integrates with Model Studio and GPU-powered Hologres AI nodes to provide a set of mainstream LLMs. Deploy a model from the console with a single click, then call it from SQL using AI Functions.
AI Functions automatically route SQL calls to the deployed model. Common functions include:
ai_embed— generate vector embeddings from text or multimodal dataai_gen— generate text using an LLM
Because the entire workflow runs in SQL, you skip the overhead of writing glue code, deploying a separate AI service, or exporting data to an external system. Storage, retrieval, and inference all happen in one place, with data and models managed together.
For more information, see AI Models and AI Functions.
Unstructured data object tables
In AI workloads, the most valuable signals often live in unstructured data—text documents, images, audio, and video. Hologres 4.0 and later supports Object Tables, which expose unstructured data and its metadata stored in an OSS data lake as queryable tables.
When combined with AI Functions, Object Tables support automatic embedding and segmentation of unstructured data. The results are stored in a structured format inside your Hologres instance, making them available for vector search and full-text search. This gives you a single system to explore, retrieve, and analyze all your data—structured, semi-structured, and unstructured.
For more information, see Unstructured Data (Object Tables).
Vector search
Hologres 4.0 includes an upgraded vector search engine built on the HGraph algorithm with hybrid memory and disk indexing. It handles trillion-scale vector datasets for both ingestion and recall, with high performance across recall speed, recall accuracy, and index build time.
Vector search integrates with AI Functions for end-to-end multimodal pipelines: transform raw data, vectorize it, ingest it, and query it—all in SQL, without deploying a separate vector database.
Use cases include:
Similarity search — find semantically related documents, products, or records
Image retrieval — search a media library by visual content
Scene recognition — classify and query visual data by detected scene
For more information, see Vector Computation Overview.
Full-text search
Hologres 4.0 uses Tantivy, a high-performance full-text search engine, with the BM25 ranking algorithm. It supports multiple tokenizers for multi-language indexing and inverted index construction.
Available search modes:
Keyword match — exact term lookup
Phrase search — match a sequence of terms in order
Natural language search — rank results by relevance to a query phrase
Term search — match documents containing a specific token
Combining full-text and vector search can boost AI performance in text-based retrieval tasks, such as retrieval-augmented generation (RAG).
For more information, see Full-text search.
MCP
The Hologres MCP Server implements the Model Context Protocol (MCP), a standard interface for connecting AI agents to data systems. Through the MCP Server, AI agents can retrieve Hologres metadata and run SQL operations.
Hologres supports multi-channel deployment of the MCP Server alongside the ChatBI Agent, giving AI applications flexible access to your data warehouse.
For more information, see MCP and Chat BI.
AI ecosystem integrations
Dify
Dify is an open-source LLM application development platform that combines Backend-as-a-Service (BaaS) and LLMOps principles. The hologres_text2data plug-in, available on the Dify official marketplace and on GitHub, lets you connect Hologres to Dify to build enterprise ChatBI applications.
For more information, see Build a ChatBI application fast using Dify and Hologres.
Python SDK
Hologres is compatible with PostgreSQL 11. Access it from Python using Psycopg for high-performance data reads and writes. For vector workloads, Hologres also provides a vector-enabled Python SDK with high-performance, low-latency vector computation support.
For more information, see Search SDK.