SynapCores v1.11.0 — Point Metabase (or any MySQL tool) at an AI-native database

Published on July 18, 2026

SynapCores v1.11.0 — Point Metabase (or any MySQL tool) at an AI-native database

Last release, SynapCores learned to speak the MySQL dialect — ON DUPLICATE KEY UPDATE, JSON accessors, ENUM, mysqldump. This release, it learned to answer the phone like MySQL.

v1.11.0 ships a MySQL wire-protocol listener. Your existing apps, IDEs, and BI tools connect to SynapCores with nothing but a connection string — no SDK, no adapter, no query rewrite. And because it's still SynapCores underneath, EMBED(), GENERATE(), and AGENT_RUN() are one query away inside every one of those tools.

What actually changed

Turn it on in gateway.toml (it's off by default, loopback-bound):

[mysql_wire]
enabled = true
bind    = "127.0.0.1"
port    = 3306

Now connect — with the same admin credentials printed at first boot, no API key to mint:

mysql -h 127.0.0.1 -u admin -p<admin-password>
mysql> SELECT name, COSINE_SIMILARITY(vec, EMBED('gym earbuds')) AS score
    -> FROM products ORDER BY score DESC LIMIT 1;
+----------+--------+
| name     | score  |
+----------+--------+
| AeroBuds | 0.83   |
+----------+--------+

Semantic search — over the MySQL wire — from the plain mysql CLI.

The tools that just work now

We validated against the drivers that matter: JDBC / Connector-J (which means Metabase, DBeaver, DataGrip), Node mysql2, PHP mysqli, mysql-connector-python, and the mysql CLI. Point Metabase at the columnstore and build dashboards on billions of rows. Browse your schema in DBeaver — SHOW CREATE TABLE, information_schema, and friends all answer like MySQL. Keep mysqldump around for egress, because lock-in is a choice and we'd rather you didn't feel any.

Why this is the interesting part

Most "connect your BI tool to a database" stories end there. This one has a twist: the database on the other end of that ordinary MySQL connection also does vector search, graph/GraphRAG, in-database AutoML, and durable AI agents — in the same engine, on the same data. You didn't add an AI database to your stack. You pointed the tools you already have at one.

The wire protocol is a translation layer over the same execution path the REST API and SDKs use, so there's one engine underneath and nothing to keep in sync. The native REST / WebSocket / SDK path is unchanged, and the listener is strictly opt-in.

Honest fine print

One known edge: mysql-connector-python's server-side prepared statements, under heavy connection concurrency, can rarely (~2%) return a spurious 1210. It's retry-safe, and JDBC/Connector-J, PHP, Node, Go, and the mysql CLI are all unaffected. We'd rather tell you than have you find it.

Try it

docker pull synapcores/community:v1.11.0-ce

Drop-in from v1.10.0, no migration. Turn on [mysql_wire], point your favorite MySQL tool at it, and run a query with EMBED() in it — then tell us which tool you connected.

Browse 187 runnable recipes → · Read the docs →