zanith

Audit + history

Four bookkeeping tables in your live database. Every migration run, every step, every snapshot is queryable like data.

The four tables

TableRows hold
_zanith_migrationsApplied migration ids, with appliedAt, appliedBy, total elapsed time, outcome (ok / failed / partial).
_zanith_migration_stepsOne row per op in each migration: kind, risk level, started-at, elapsed-ms, outcome, error message if any.
_zanith_schema_snapshotsSerialized schema graph at each migration boundary. phase is before / after / baseline.
_zanith_artifactsSoft-drops, archives, backfill checkpoints — see [Recovery](/docs/migrate/recover).

All four are normal Postgres tables. Query them, alert on them, ship them to your warehouse. The schema is stable across engine versions.

The history CLI

SHELL
zanith migrate history                 # most-recent-first list of applied migrations
zanith migrate history --verbose       # include per-step outcomes + risk
 
zanith migrate history --since 7d      # last week
zanith migrate history --json          # machine-readable
 
zanith migrate history --id 20260504_120000_add_orders_index
# → full detail for one migration: every step, snapshot diff, artifacts created

Snapshot replay

Each successful migration writes a serialized graph snapshot. That gives you time-travel for the schema — load a past snapshot, diff it against current, see what changed.

TS
import {
  findSnapshot,
  deserializeSchemaGraph,
  diffSchemas,
} from 'zanith';
 
// Schema graph as it was after a specific migration.
const row = await findSnapshot(adapter, '20260420_120000_drop_orders_total');
const past = deserializeSchemaGraph(row!.serialized);
 
// Diff against the current declared graph.
const drift = diffSchemas(past, currentGraph);
console.log(`${drift.length} structural changes since that migration`);

Querying like data

SQL
-- Migrations slower than 30 seconds, in the last week.
SELECT id, applied_at, elapsed_ms / 1000.0 AS seconds
FROM _zanith_migrations
WHERE applied_at > now() - interval '7 days'
  AND elapsed_ms > 30000
ORDER BY elapsed_ms DESC;
 
-- Failed steps with their error messages.
SELECT migration_id, step_index, op_kind, error
FROM _zanith_migration_steps
WHERE outcome = 'failed'
ORDER BY started_at DESC
LIMIT 50;
 
-- Total destructive ops applied this month.
SELECT count(*)
FROM _zanith_migration_steps
WHERE risk_level >= 4
  AND started_at >= date_trunc('month', now());

Programmatic reads

TS
import { listAppliedMigrations, listMigrationSteps } from 'zanith';
 
const all = await listAppliedMigrations(adapter);
const steps = await listMigrationSteps(adapter, all[0].id);

Same shapes as the SQL above, plus type safety. See programmatic API for the full surface.