Oracle gave a public status update on November 16, 2026 about native vector index support in MySQL. In the second post of a series on open MySQL development, MySQL Community Architect Henrik Ingo described a joint project to add a VECTOR index type to the InnoDB storage engine.
The update matters because InnoDB does not yet have a native index for fast vector search, and this project is how MySQL plans to add one. Oracle, Google, AWS, Intel and Percona are all named as taking part, and the design is being discussed in the open on GitHub.
What Oracle Announced
The post does not announce a release. It reports where the work stands and which options are still open. The main points are:
- The contributors have agreed to build a native index type in the MySQL core, inside InnoDB.
- The current focus is ScaNN, a library that Google's Cloud SQL team recently open sourced. A pull request based on it has been opened, and Oracle's InnoDB team is integrating and benchmarking it.
- Intel has proposed its Vamana algorithm as another candidate.
- AWS engineer Hugo Wen started the project with a talk at the Contributor Summit in May.
- Intel and Percona are presenting further contributions at the Contributor Summit held this week.
Ingo adds that the ScaNN code will only be merged after review, benchmarking and comparison with the alternatives. Nothing in the post gives a target version or date for the feature.
Four Algorithms on the Table
Vector indexes answer approximate nearest neighbor (ANN) queries: find the rows whose vectors lie closest to a given vector, accepting a small loss of accuracy for a large gain in speed. The post names four candidate algorithms, which fall into two families.
- HNSW keeps vectors in a layered graph and walks from sparse upper layers down to the full set. It is the usual baseline and is found in pgvector, Lucene, ClickHouse and DuckDB, but it works best when the index fits in memory.
- IVF first groups vectors into clusters, then searches only the cluster nearest to the query. It is compact, but results can be slightly off when a query falls near the border of a cluster.
- ScaNN belongs to the IVF side and concentrates on compressing vectors with as little error as possible. The post cites reports of speed gains of four to eight times in benchmarks.
- Vamana is a graph index like HNSW.
The team has not decided whether to benchmark the four and keep one, or to ship a pluggable interface that lets several libraries be used. Ingo says MySQL may end up including all four.
Open Design Questions
Algorithm choice is only half of the work. A second debate is about storage: the index could live in a separate InnoDB table linked to the base table, called a sub-table, or it could be built as a regular secondary index using InnoDB's own page layout.
The secondary index route looks cleaner, the post notes, but would need far more work in sensitive areas such as locking and row visibility.
The SQL syntax is also unsettled. Tickets on GitHub discuss how CREATE VECTOR INDEX should look and how queries should read from the index. Cosine distance is confirmed as one of the distance functions that will be implemented.
The long-term goal is hybrid search, where a vector condition is combined with ordinary predicates and the optimizer weighs the vector index against other indexes on the table.
MyVector Is Available Today
While the native index is being built, the post points to MyVector, a plugin and user-defined function package written by Alkin Tezuysal. It supports HNSW and IVF, exact and approximate search, and several distance functions. It keeps its indexes current by watching the binary log, and it has been in production use since March.
Its advantage is simple: it can be downloaded and tried now. Ingo also lists other community projects, including mysql_vss, ng_vector_search and the ClawDB storage engine.
Wider Changes in MySQL Development
The update also covers the project itself. Oracle is getting the mysql-server repository on GitHub ready to become the upstream for all open source MySQL work, which reverses the direction in which patches used to flow between editions.
MySQL Test Run, a code formatter and the Oracle Contributor Agreement bot are now wired into that repository. A weekly newsletter lists every issue and pull request that saw activity. Engineers are working toward MySQL 9.7 LTS and the MySQL 26.7 innovation release, with more attention on Community Edition.
Outside contributions are already landing in this model, as with the BOLT binary log optimization that AWS contributed to MySQL earlier this month.
What It Means for Developers
If you need vector search on MySQL in production right now, do not wait for the native index. Use a plugin such as MyVector, or keep embeddings in a system that already indexes them. Be ready to migrate later, because the final syntax and index type are not fixed.
If you are planning for next year, this is a good time to follow the GitHub discussions. Teams with real workloads can influence the choice by sharing data sizes, dimension counts and recall needs while benchmarks are still being run.
The memory profile will matter for capacity planning. A graph index that must fit in RAM sizes very differently from a clustered, compressed one, so the algorithm that wins will shape instance choices. Other engines are moving fast here too, as the recent MongoDB 9.0 vector search speedup shows.
Conclusion
MySQL does not have a native vector index yet, but the work is real, public and backed by several large vendors. The ScaNN pull request is the furthest along, and the choice between one algorithm and a pluggable design is still open. The next post in Oracle's series will cover the MySQL Component Framework and extensibility.
Sources
See my articles more often when you search Google.
