Export data from LangSmith

1. Set the archive scope

Record the project, environment, start time, exclusive end time and filters. Include child runs if you need complete traces.

• Preserve run, trace and parent IDs plus timestamps.
• Handle prompts, datasets, scores and permissions separately.
• Use a new file for each window to avoid duplicate appends.

2. Export a sample

Configure the LangSmith credentials and endpoint, then set ARCHIVE_PROJECT. This JSONL example includes child runs and applies the upper date boundary locally; it is not a resumable whole-history exporter.

import json
import os
from datetime import datetime, timezone
from langsmith import Client

start = datetime(2026, 9, 1, tzinfo=timezone.utc)
end = datetime(2026, 9, 2, tzinfo=timezone.utc)
client = Client()
count = 0

# Exclusive creation protects an existing archive from replacement.
with open("langsmith-sample.jsonl", "x", encoding="utf-8") as archive:
    for run in client.list_runs(
        project_name=os.environ["ARCHIVE_PROJECT"],
        start_time=start,
    ):
        if not start <= run.start_time < end:
            continue
        archive.write(json.dumps(run.dict(), default=str) + "\n")
        count += 1

print(f"Archived {count} runs; reconcile with the source project.")

3. Scale with bulk export

Eligible accounts can export Parquet to S3-compatible storage. Check plan and signup-date requirements, set a fixed interval, and wait for the job’s completed status.

4. Validate and prepare the import

Reconcile unique run IDs against the source filters and inspect a nested failure. Keep checksums and the query beside the archive.

• Record parents outside the export window.
• Map selected inputs and expected outputs to Lunary’s CSV/JSONL dataset format.
• Retain a dataset-row-to-source-ID map; raw LangSmith runs are not native Lunary traces.