Export data from Langfuse

1. Export a filtered sample

Set the project, time window, environment and tags, then export CSV or JSON from the relevant table. Hidden UI columns are still exported; inspect the file before sharing it.

• Record filters and timezone.
• Confirm the exported object type and schema.
• Handle prompts, scores and datasets as distinct objects.

2. Choose a repeatable export route

For larger archives, check supported objects and destination settings for blob-storage or programmatic exports. Merge fixed intervals by stable record ID so retries do not duplicate records.

3. Reconcile the archive

Compare unique IDs, nested relationships and scores with the source. This manifest is an audit template, not executable export configuration; fill counts after verification and store file checksums separately.

{
  "source": "langfuse",
  "project": "your-project",
  "startInclusive": "2026-09-01T00:00:00Z",
  "endExclusive": "2026-09-02T00:00:00Z",
  "filters": { "environment": "production" },
  "exportRoute": "ui-json",
  "exportedObjectType": "record-the-actual-type",
  "sourceCount": null,
  "uniqueExportedCount": null,
  "missingRelationships": null,
  "verifiedAt": null
}

4. Prepare selected examples for Lunary

Keep the raw files unchanged. Map inputs, observed outputs and approved expected answers to Lunary’s CSV/JSONL dataset format; this does not recreate native trace history.

• Keep a row-to-source-ID map.
• Verify imported counts and representative examples.
• Check new instrumentation before retiring the old integration.