Product overview

pocketPG is a PostgreSQL observability and tuning workspace. You connect an instance, and pocketPG reads live system catalog data to surface health signals, query behavior, bloat, tuning recommendations, and operational risks.

How it works

1. Connect — paste a PostgreSQL URL or configure an SSH tunnel. Credentials are stored server-side for saved instances.
2. Observe — the Overview tab aggregates findings across health, queries, indexes, and bloat.
3. Diagnose — drill into specific tabs (Health, Queries, Indexes, Vacuum, Activity, WAL) to isolate the source of pressure.
4. Tune — review recommendations with evidence, confidence scores, and rollback SQL. Copy the command or provider CLI.
5. Maintain — run guarded VACUUM/ANALYZE/REINDEX, enable webhook alerts, and set up daily digest emails.

What pocketPG reads

All data comes from PostgreSQL system views and functions:

• pg_stat_database, pg_stat_user_tables, pg_stat_user_indexes
• pg_stat_statements (when installed)
• pg_stat_activity, pg_stat_bgwriter, pg_stat_wal (PG14+)
• pg_stat_progress_vacuum
• pg_settings, pg_file_settings
• pg_catalog views for sequences, constraints, extensions

pocketPG never creates tables, indexes, extensions, or functions on your database.

Safety model

• Read-only — all connections use autocommit = True so monitoring queries never appear as open transactions.
• Statement timeout — monitoring queries run with a timeout so one slow diagnostic query does not stall the app.
• No schema changes — pocketPG reads system views only. The one exception is owner-initiated maintenance (VACUUM, ANALYZE, REINDEX) which is gated behind confirmation modals and role checks.

What makes it different

	pocketPG	Agent-based tools (pganalyze, Datadog)
Setup	Connect and go — no agent, no binary, no schema changes	Install agent on host, configure credentials, wait for data collection
Architecture	On-demand queries per tab load, 30s cache	Continuous background polling
Tuning	Evidence-based recommendations with rollback SQL	Varies — often index-only or alert-only
Sharing	Owner-pays model — invite team members at no extra cost	Per-seat or per-server pricing
Deployment	SaaS on Railway — nothing to self-host	Agent + SaaS or self-hosted