Your cluster context, not generic advice

We answer from your data.

‍

What makes the difference

•      Reads live state before answering. Queries pg_stat_user_tables, pg_stat_replication, and slow query logs in real time. The answer reflects what is happening now.

•      Compressed context, not raw logs. The AI receives a structured summary — top queries, error clusters, metric anomalies — not gigabytes of raw log lines.

•      Explains before it acts. Every recommendation includes the reasoning. You see which metric triggered the suggestion and what changes.

‍

COST-EFFICIENT AI

No hidden API costs. No per-query billing.

The reason most AI database tools get expensive fast: they send raw logs to an external LLM API and charge per token. 123Cluster does not work that way.

SECURITY-FIRST AI

Your data stays inside your boundary

Infrastructure data — query patterns, schema shapes, connection strings — is sensitive. The AI operates inside your deployment, not in a shared SaaS layer.

‍

‍

EXAMPLES

What it looks like in practice

Log analysis — missing index after deploy

you:  What's killing query performance since the deploy at 14:30?

AI:   Top offender — sequential scan on orders.customer_id (rank #1 by total time).

   Appeared 4,847 times since 14:30. Avg 2.3s, max 18.4s.

   The query fingerprint matches your new reporting endpoint.

   There is no index on customer_id — the column was added in today's

   migration but the index was not. Table has 47M rows.

   Fix: CREATE INDEX CONCURRENTLY idx_orders_customer_id ON orders(customer_id);

   Non-blocking. Estimated build time: ~12 minutes at current write rate.

Parameter advisor — autovacuum on high-write table

you:  autovacuum keeps falling behind on the events table

AI:   Confirmed. events has 8.2M dead tuples — 16x over trigger threshold.

   Last vacuum: 3 days ago. Current write rate: ~9,400 updates/sec.

   Root cause: autovacuum_vacuum_scale_factor = 0.2 (default).

   At this write volume the trigger fires once every ~3 days. Should be hourly.

   Recommendation (table-level, non-global):

   ALTER TABLE events SET (autovacuum_vacuum_scale_factor = 0.01,

                             autovacuum_vacuum_threshold = 1000);

   Apply during low-traffic window. Run VACUUM ANALYZE events after.

Trend detection — replication lag alert

ALERT: Replication lag trend — eu-west-2b

AI:   Lag has grown from 0.3s to 4.8s over the past 40 minutes.

   Rate of growth is accelerating, not plateauing.

   Correlation: WAL generation spike at 06:14 UTC (+320% vs 7-day avg).

   The replica is not I/O bound — walreceiver is the bottleneck.

   Most likely cause: bulk write job started at 06:10.

   At current rate, lag will exceed 30s in ~18 minutes.

   Options: set synchronous_commit=local for the session,

   or redirect reads to eu-west-2a during the write window.

  Connect a cluster and ask the first question.

  No setup. The AI is live the moment your cluster is connected.

‍