Postgres sink (production)

import { postgresSink } from "llmmeter-postgres";

const sink = postgresSink({
  connectionString: process.env.LLMMETER_DB_URL,
  batchSize: 50,
  flushIntervalMs: 1000,
});

The Postgres sink uses the pg connection pool. On first write it bootstraps the llmmeter_calls table with appropriate indices (override with skipSchemaInit: true if you manage migrations).

Read API mirrors the SQLite read API exactly so the dashboard works without code changes:

• listCalls(pool, filters)
• aggregateByBucket(pool, bucketSeconds, filters)
• topByDimension(pool, dim, filters)
• totals(pool, filters)
• pruneOlderThan(pool, beforeTs)

Use this for any deployment where multiple processes need to write to the same store, or where you want to retain millions of records.