SuperAPI AI Directory
Partner with us中文AI newsFavorites
DirectoryNews and tutorials

Hologres - A one-stop Agentic multimodal retrieval and analytics platform launched by Alibaba Cloud

Author: AI小集 · Original source: AI工具集

What is Hologres

Hologres is launched by Alibaba Cloud as aone-stop Agentic multimodal retrieval and analytics platform, compatible with the PostgreSQL protocol, supporting PB-level real-time writes, updates, OLAP/ad hoc analytics, and high-concurrency point queries. The platform uses unified SQL to connect structured, semi-structured, and OSS unstructured data. Combined with Object Table, Dynamic Table, and AI Function, it provides full-text, vector, and hybrid retrieval as well as model invocation capabilities, and extends to Agent long-term memory and an observability foundation, making it suitable for Data+AI/RAG scenarios.

Hologres

Main features of Hologres

  • Real-time data warehouse OLAP: Supports PB-level real-time data writes, updates, multidimensional analysis, and ad hoc queries.
  • High-concurrency point query Serving: Compatible with the PostgreSQL protocol, designed for low-latency online queries and high-concurrency data services.
  • Full-text/vector/hybrid retrieval: A single copy of data supports full-text retrieval, vector retrieval, hybrid retrieval, and RAG recall.
  • Multimodal Data+AI processing: Use Object Table to connect to OSS unstructured files, Dynamic Table for incremental processing, and AI Function for model invocation such as parsing, chunking, Embedding, translation, and generation.
  • Lakehouse and multi-source integration: Can connect to MaxCompute, Paimon, Iceberg, Flink, DataWorks, and external data sources such as MySQL/PostgreSQL.
  • Data warehouse AI assistant: Encapsulates experience such as diagnostics, SQL generation, specification recommendation, and cost governance into executable Skills.

How to use Hologres

  • Activate a trial: Go to the Alibaba Cloud Hologres official website https://cn.aliyun.com/product/hologres，点击开通/产品试用并完成账号与计费确认。
  • Create an instance: Create a Hologres instance and select the region, network type, CU specification, and storage method.
  • Configure access: Configure VPC/public network, IP whitelist, database accounts, and least-privilege permissions.
  • Connect to the instance: Use HoloWeb, psql, JDBC/ODBC, or BI tools to connect to the instance and create a database.
  • Model and create tables: Create tables based on the workload, distinguish between row storage/column storage, and add primary keys, distribution keys, vector indexes, and full-text indexes as needed.
  • Ingest data: Ingest data into Hologres through INSERT/COPY, Flink, DataWorks, MaxCompute, or synchronization tools.
  • Process multimodal data: Create Object Table to map OSS files, use Dynamic Table for incremental processing on the lake, and use AI Function to complete parsing, chunking, Embedding, or generation.
  • Unified query and analysis: Use unified SQL to execute point queries, OLAP, full-text/vector/hybrid retrieval, or AI inference, and save commonly used views and materialized queries.
  • Stress testing and launch: Before launch, conduct stress testing on concurrency, latency, recall accuracy, import throughput, and model Token cost.
  • Operations optimization: After production operation begins, enable monitoring for slow queries, resource consumption, cost, and auditing, and elastically scale up or down as needed.

Core advantages of Hologres

  • Integrated architecture: Converges OLAP, point queries, full-text, vector, hybrid retrieval, and AI inference into the same engine, reducing assembly of multiple components such as ClickHouse/Greenplum + HBase/Redis + Elasticsearch.
  • One dataset reused across multiple scenarios: The same dataset simultaneously serves BI reports, ad hoc analysis, online services, and RAG/Agent retrieval, reducing sync pipelines and the risk of data inconsistency.
  • Strong real-time capabilities: Supports real-time writes, real-time updates, and query-after-write, making it suitable for scenarios where high-concurrency, low-latency Serving and real-time analytics coexist.
  • Data+AI native: Through Object Table, Dynamic Table, and AI Function, OSS files, incremental processing on the lake, and model calls are brought into the SQL pipeline.
  • Good integration with the lakehouse and Alibaba Cloud ecosystem: Closely connected with components such as MaxCompute, Flink, DataWorks, OSS, and Bailian, making it suitable for implementation within the Alibaba Cloud ecosystem.
  • Fully managed and elastic: Cloud-native managed service reduces O&M costs and can elastically adjust resources based on load, making it suitable for businesses with large fluctuations in data volume and concurrency.

Comparison of Hologres with similar competing products

Dimension Hologres ClickHouse
Product positioning Alibaba Cloud's integrated real-time lakehouse platform, covering OLAP, point queries, full-text/vector/hybrid retrieval, and AI inference Open-source columnar OLAP database, strong in high-speed scanning, aggregation analysis, and compressed storage
Open source and controllability Closed-source commercial cloud product, weak controllability, but no O&M required Open-source core, can be self-built or use ClickHouse Cloud/Altinity, strong controllability
SQL/protocol Compatible with the PostgreSQL protocol, friendly for BI/business integration Strong SQL-like capabilities, but the dialect and ecosystem tools are not fully consistent with PG/MySQL
Real-time writes/updates Focuses on real-time writes, real-time updates, and query-after-write Strong append writes; high-frequency updates/real-time primary key updates require versioning capabilities and merge strategy support
Point query Serving Natively oriented toward high-concurrency, low-latency online services Can perform point queries, but its advantages lie more in large-batch scan aggregation
OLAP performance Balances real-time OLAP and online Serving Usually stronger in single-query throughput, columnar compression, and wide-table aggregation
Full-text retrieval Built-in full-text retrieval, can be combined with vector/scalar hybrid recall Has full-text capabilities, but search relevance and tokenization ecosystem are usually not as good as dedicated retrieval systems
Vector/hybrid retrieval Native support for vector retrieval and hybrid retrieval, close to RAG Vector capabilities are available, but often require additional indexes/components and tuning
Data+AI pipeline Object Table + Dynamic Table + AI Function, AI processing can be brought into SQL Usually externally connects ETL, model services, vector databases/search systems, flexible but fragmented

Hologres application scenarios

  • Real-time BI and business analysis: Writes transaction, order, and traffic data in real time, uniformly supporting report dashboards, ad hoc analysis, and business monitoring.
  • Real-time risk control and marketing: Combines real-time events, user profiles, and vector/full-text recall to perform millisecond-level risk determination, similar case retrieval, and personalized recommendations.
  • Enterprise RAG knowledge base: After parsing PDFs/images/audio and video, it chunks and embeds them, using hybrid retrieval to serve Q&A, customer service assistance, and policy retrieval.
  • Customer service voice and ticket analysis: Performs speaker diarization, ASR transcription, sentiment analysis, and summary generation on recordings, turning them into searchable data assets.
  • Agent long-term memory and multimodal inspection: Provides cross-session memory, image comparison, and discrepancy verification capabilities for scenarios such as AI learning companions, AI Coding, and shelf inspection.