SynapCores

SynapCores v2.4.6-ce: Graph Explorer restored, plus a field-report fix batch

Published on November 5, 2026

SynapCores Community Edition v2.4.6-ce

Community Edition · free · seamless upgrade from v2.4.x

v2.4.6-ce puts the Graph Explorer back to one-click working on real result sets, gives AutoML a clean teardown statement, and clears a batch of correctness and speed issues surfaced by a field regression report — across SQL, the data lake, and the in-database agent. Everything here is drop-in: no data migration, no config change, existing data read as-is.

Explore any graph query in the UI without it falling over

Running a modern MATCH in the Graph Explorer used to blank out with "Something went wrong" on every non-empty result. It now renders results reliably, and the profile view degrades gracefully instead of erroring. The graph view is usable again on large, real graphs — not just toy queries.

Drop an AutoML experiment and everything it produced, in one statement

New DROP EXPERIMENT [IF EXISTS] <name> removes the experiment, its trial history, and the model it trained — and refuses, naming the blocker, if a separately deployed model still depends on it, so you never orphan a live model by accident. It's documented in both the runtime SQL manual (what your MCP clients introspect) and the public SQL reference.

Trust ANOMALY_SCORE() to stay fast when you wrap it

Wrapping anomaly scores in ROUND(...) or a similar expression no longer drops the optimized batch path — a scored column computes once per batch instead of once per row. Dashboards and alerting queries over large tables stay fast.

Longer-running in-database agents finish instead of timing out

AGENT_RUN work that legitimately needs time now completes within its budget and releases cleanly when a deadline is hit, instead of stalling. The learning-loop goal view refreshes on its own as steps advance.

Also fixed, from the field regression report

  • MySQL-compatible DATE_SUB / DATE_ADD two-argument INTERVAL form, and GROUP_CONCAT / STRING_AGG with SEPARATOR.
  • Immutable tables write and verify against the named database you're in, not an ambient default.
  • COPY TO LAKE partitioning writes the expected number of files without spurious budget errors.
  • Natural-language questions resolve "this month" / "last month" to the correct dates, and database-scoped API keys answer over exactly their granted tables.
  • MCP execute / query accept an optional database argument.
  • Cleaner operational logs — no more error-level noise on an ordinary DROP TABLE.

Proven before shipping

Zero regressions versus the previous release across the full recipe-certification sweep (identical pass set), the state-asserting feature validator fully green, and a clean boot plus model run on a non-AVX-512 CPU. Drop-in upgrade: no on-disk format change, and the storage back-compat gate passed.

Get it

docker pull synapcores/community:v2.4.6-ce

Or upgrade in place — existing data is read as-is.

LM