deep-Postgres performance coach · 17 advisor cards · private beta
The AI Postgres performance coach
built for DBAs who know the internals.
Connect a read-only role. Paste an EXPLAIN or slow log. Get prioritized findings across workload, schema, maintenance, and cluster — with the exact DDL, the HypoPG plan diff, and a pull request in your migration tree. No auto-apply.
Both free, no login. EXPLAIN Explainer walks you through a plan; demo shows the 17-card dashboard with sandbox data.
Private beta · join the waitlist
Continuous monitoring rolls out by invite. $29/mo at general availability.
Why this exists
Slow queries kill product velocity.
The 200ms creep to 2s happens in production, not staging. By the time a DBA sees it, your P95 already tanked.
DBAs are rare and expensive.
Most teams under 100 engineers don't have one. The ones that do are juggling six databases each.
Existing tools surface, not solve.
pganalyze flags symptoms but doesn't recommend DDL. EverSQL got acquired. OtterTune shut down. SolarWinds DPA isn't AI-native. DBAs end up doing the same root-cause walk every time.
How TuneStack is different
Postgres-internals depth.
Reads pg_stat_statements, pg_stat_activity, pg_stats, replication slots, and TOAST storage. Simulates indexes via HypoPG with EXPLAIN cost diffs before recommending.
DBA-grade reasoning, not pattern matching.
Agentic LLM loop with tools: identify the slow node, simulate the index, validate the rewrite with sqlglot, score the cost delta. Nothing is surfaced without a verifiable plan-cost win.
Human-in-the-loop, not auto-apply.
OtterTune's tombstone: never auto-apply. We generate the DDL, show the plan diff, and open a pull request in your migration tree. You review. You merge.
Who it's for
Backend engineers without a DBA.
You read EXPLAIN occasionally and Google what “sequential scan” means. TuneStack turns a slow query into a walkthrough, a fix, and a pull request.
DBAs and architects at larger shops.
You know what's wrong — you just want the 60-minute investigation done in 60 seconds, and the fix to land in the migration tree, not a Slack DM.
Everything you get
17 advisor cards on every dashboard. Each one a real Postgres internal, not a generic metric.
Workload
LockWait
top blocked/blocking pairs across ticks
CallSiteAttribution
sqlcommenter tags → where in your code the slow query lives
Regression
fingerprint-level latency deltas vs baseline
PlanFlip
structural plan changes between snapshots
WorkloadTrend
sustained latency-slope detection
Schema + indexes
SchemaAudit
unused indexes, missing FKs, denormalized hot spots
IndexAdvisor
HypoPG-simulated workload_benefit_pct ranking
Partitioning
range/list/hash recommendation with key + window
SchemaDrift
table/column/index churn between snapshots
Maintenance
Maintenance
vacuum, ANALYZE, REINDEX prioritized by impact
Autovacuum
never-vacuumed / starved / disabled-per-table
TableBloat
pg_repack-grade bloat ratios with reclaim estimate
IndexBloat
REINDEX CONCURRENTLY candidates ranked
Toast
EXTERNAL storage + column-split candidates
Cluster
ConnectionPool
idle-in-xact leaks, saturation, pgbouncer mode advice
Replication
replay/write lag, inactive slot WAL retention
CollectorHealth
tick freshness, cert binding, agent error feed
/explain
EXPLAIN Explainer
Paste one EXPLAIN plan. Get a plain-English walkthrough, the bottleneck nodes, indexes, and rewrites.
Free, no login →
/slow-log
Slow-log Analyzer
Paste a Postgres slow log or pg_stat_statements CSV. Get top-N ranked across CPU, I/O, memory, and network — with a plain-English finding per axis.
Free, no login →
/dashboard
Continuous monitoring dashboard
Deploy the collector in your VPC. Get regression alerts, HypoPG index cost diffs, schema audit, and maintenance advisor — updated every snapshot.
$29/mo →
Learn
Start free. Get on the waitlist for production.
Free tools need no login — paste in, get a result, share it. Continuous monitoring is private beta; first cohort onboarding now. Drop your email and we'll get in touch when your slot opens.