MyScale
Officially listedSQL vector database built on ClickHouse, unifying vector search, full-text search, and SQL.
MyScale
MyScale is a SQL vector database built on open-source ClickHouse, unifying SQL database, vector database, and full-text search in a single system. It lets SQL-savvy developers build AI applications without learning new APIs, and its MSTG vector engine delivers a 10x boost in data density, optimized for RAG retrieval and vector search scenarios.
Core Capabilities
- SQL-native queries: Vector search using standard SQL syntax with zero learning cost, supporting complex WHERE clause filtering
- MSTG vector engine: A proprietary vector search engine leveraging NVMe SSDs for 10x data density, with a single Pod storing 10 million 768-dimensional vectors
- Unified data platform: Manages structured data, vector data, full-text search, and JSON data together, cutting infrastructure complexity by 60%
- High-performance filtered search: A columnar storage architecture delivers 10x better filtering performance than row-based databases, supporting joint SQL WHERE and vector search queries
- Multimodal data support: Handles numeric, datetime, text, geospatial, JSON, vector, and time-series data types
- Mainstream framework integrations: Deep integration with LangChain, LlamaIndex, OpenAI, and Dify, with Python, Node.js, and Go clients
Use Cases RAG knowledge bases, enterprise Q&A systems, academic research platforms, legal document review, financial analysis, customer support, image search, AI chatbots, and other complex AI applications that must handle both structured and vector data — especially retrieval scenarios demanding high-precision metadata filtering.
Unique Advantages While dedicated vector databases like Pinecone and Milvus require learning new APIs, MyScale lowers the learning curve with SQL. Official figures show query speeds 10x faster than Pinecone s1 Pods, 3.6x better cost-performance than dedicated databases, and 60% lower infrastructure costs. Built on the mature ClickHouse stack, stability is assured.
Editor's Review The technical architecture is solid — built on battle-tested ClickHouse — and the MSTG vector engine's performance claims are attractive. The free tier supports 5 million vectors and paid plans start at $68/month, friendly to cost-sensitive teams. The only concern is a 0.4% market share indicating an early-stage product with relatively limited user case studies and community resources. SQL-fluent dev teams are advised to start evaluating on the free tier to verify it meets performance needs.
Pricing
### 💰 定价模式：免费增值 **起步价**：$0（永久免费层） #### 主要方案 - **Development（免费）**：$0/月 - 支持 500 万个 768 维向量，单配置单副本，社区支持 - **Standard（标准版）**：$68/月起 - 多配置多副本多可用区，容量优化 Pod（双倍存储，成本降 43%），工作时间邮件支持含 SLA - **Enterprise（企业版）**：定制报价 - 完全定制化配置，99.9% 正常运行时间保证，7×24 小时邮件支持，专属技术支持 #### 试用/其他信息 基于 Pod 使用时长计费，提供在线价格估算工具。单个 x1 Pod 可存储 1000 万个 768 维向量。 — Visit website
FAQ
Is MyScale free?
It offers a permanent free tier supporting up to 5 million 768-dimensional vectors, with community support. Paid plans start at $68/month with more configuration options and SLA guarantees.
What are MyScale's main features?
Core features include SQL-native vector search, the MSTG high-performance vector engine, unified management of structured and vector data, full-text search, hybrid search, SQL WHERE filtered queries, multimodal data support, and LangChain/LlamaIndex integrations.
Is MyScale good for building RAG applications?
Very much so. MyScale's ability to run SQL WHERE clauses jointly with vector search excels in RAG scenarios that need complex metadata filtering, with significantly improved retrieval precision.
How does MyScale differ from Pinecone?
MyScale uses standard SQL instead of a proprietary API, so the learning cost is lower; it also supports unified management of structured and vector data. Official figures show queries run 10x faster than Pinecone s1 at lower cost.
How many vectors can a single MyScale instance store?
The free tier Pod supports up to 5 million 768-dimensional vectors. A single x1 Pod on the standard edition stores 10 million 768-dimensional vectors, and capacity-optimized Pods offer double the storage capacity at 43% lower cost.
Which programming languages and frameworks does MyScale support?
It offers Python, Node.js, and Go clients plus JDBC and HTTPS APIs, with deep integrations for mainstream LLM frameworks including LangChain (Python & JS/TS), LlamaIndex, OpenAI, and Dify.
What are the advantages of MyScale's MSTG vector engine?
MSTG (Multi-Scale Tree Graph) is MyScale's in-house vector engine, leveraging NVMe SSDs for a 10x boost in data density and outperforming dedicated vector databases on performance and cost efficiency by 4-10x.