SQLv2 is a next-generation standard that unifies traditional SQL, machine learning, vector search, generative AI, and multimedia support into a single, cohesive query language.
Built for the AI era, SQLv2 eliminates the complexity of stitching together multiple systems — relational databases, vector databases, ETL pipelines, and external ML services — by bringing everything natively inside SQL.
Modern AI applications require developers to maintain disconnected data stacks:
- A relational database for structured data
- A separate vector database for embeddings
- External machine learning APIs
- ETL pipelines to move data between systems
- Ah! And multiple GenAI endpoints
This leads to:
- Data silos and inconsistent state
- High infrastructure costs
- Latency and performance bottlenecks
- Slower development cycles
SQLv2 solves these problems by introducing AI-native primitives directly into the SQL language.
SQLv2 builds on ANSI SQL while adding new constructs designed for modern AI workloads:
| Category | SQLv2 Features |
|---|---|
| AI/ML Functions | PREDICT(), CLASSIFY(), GENERATE(), SENTIMENT() |
| Vector Search | Native vector type (VECTOR(n)), similarity operators <=> and <->, functions like VECTOR_SEARCH() |
| Generative AI | GENERATE() for text synthesis, summarization, and content generation directly in SQL |
| Multimedia Support | First-class data types: IMAGE, VIDEO, AUDIO, PDF |
| AutoML | CREATE EXPERIMENT, DEPLOY MODEL, PREDICT ... USING ... |
| Zero-Copy & GPU Acceleration | Built-in performance for ML and vector operations |
SELECT user_id,
PREDICT(purchase_likelihood)
FROM users;SELECT product_id,
GENERATE('Write a 2-sentence description for ' || product_name)
FROM products;SELECT doc_id, content
FROM documents
ORDER BY embedding <=> EMBED('healthy recipes')
LIMIT 5;SELECT video_id,
TRANSCRIBE(video_data) AS transcript,
SUMMARIZE(TRANSCRIBE(video_data), 100) AS summary
FROM videos
WHERE JSON_EXTRACT(DIMENSIONS(video_data), '$.width') >= 1920;This repository is the official home of the SQLv2 specification. It serves as a living document and collaboration hub for the SQLv2 community.
- 📄 SQLv2 Specification – The full technical definition of the language.
- 💬 Community Proposals – Suggest new features or changes to the standard.
- 🗂 Reference Examples – Example SQLv2 queries for developers and implementers.
- 📈 Version History – Transparent changelog for all revisions of the spec.
sqlv2/
├── README.md # Introduction and overview (this file)
├── SQLv2Specs.md # Full technical specification
├── proposals/ # Community-driven feature proposals
│ └── TEMPLATE.md # Template for new proposals
├── examples/ # Example SQLv2 queries
│ ├── ai_queries.sql
│ ├── vector_search.sql
│ └── multimedia.sql
├── CONTRIBUTING.md # Guidelines for proposing changes
└── CHANGELOG.md # Version history and spec updates
We welcome contributions from database engineers, ML practitioners, and anyone interested in shaping the future of AI-native data systems.
There are three main ways to participate:
| Contribution Type | Description | Where |
|---|---|---|
| Clarification Request | Ask questions about parts of the spec that need more detail. | GitHub Issues |
| New Feature Proposal | Suggest new functions, data types, or language extensions. | GitHub Issues → Pull Request to /proposals/ |
| Implementation Feedback | Share insights from real-world usage of SQLv2. | GitHub Issues |
See our CONTRIBUTING.md for step-by-step instructions.
-
SQLv2 follows a semantic versioning approach:
- v2.0.0 – Initial stable release
- v2.1.0 – New features added
- v2.2.0 – Improvements and clarifications
-
All changes are documented in CHANGELOG.md.
-
SQLv2 was initially developed by Luis B. Mata and the SynapCores team.
-
SynapCores is the first implementation, but SQLv2 is intended to be open and vendor-neutral.
-
Decisions about the spec are made openly, with feedback from the community via:
- GitHub Issues
- Pull Requests
- Quarterly roadmap reviews
The following milestones are planned for SQLv2 evolution:
| Milestone | Status |
|---|---|
| Core spec publication | ✅ Completed |
| Public GitHub repo launch | ✅ Completed |
| Beta implementation via SynapCores | ✅ Completed |
| RFC submission to IETF | Planned |
| Multimedia deep learning extensions | Planned |
| ISO SQL committee engagement | Future milestone |
The SQLv2 specification is free to use and implement under the CC BY 4.0 license.
You are free to copy, modify, and distribute the specification, provided attribution is given to the original author(s).
The SynapCores implementation of SQLv2 is proprietary and separate from this specification.
SQLv2 was created by Luis B. Mata with the support of the SynapCores engineering team and early design partners. Special thanks to the developer community for shaping the future of this open standard.
- Clone this repository and read the SQLv2Specs.md file.
- Submit proposals for new features under
/proposals/. - Sign up for the SynapCores beta to experience the first implementation.
SQLv2 is more than a database extension — it’s the unified query language for the AI era. By bringing structured data, AI, and multimedia together, SQLv2 aims to do for AI-powered applications what SQL did for relational data in the 1970s.
Join us in shaping the future of data. Start here: github.com/synapcores/sqlv2