Importer

Langfuse

Langfuse already stores every trace and generation with its model, token usage and cost. Export the observations with your own keys and import them.

Create or use a Langfuse API key pair for the project.
Download observations page by page with the public API (below).
Run import-langfuse. Observation names are matched to the blueprint's step names.

Export and import

mkdir -p lf; for page in 1 2 3 4 5; do
  curl -s -u "$LANGFUSE_PUBLIC_KEY:$LANGFUSE_SECRET_KEY" \
    "$LANGFUSE_HOST/api/public/observations?limit=100&page=$page" > lf/obs-$page.json
done
python harness.py import-langfuse lf/*.json --blueprint .harness/harness.yaml --out .harness/runs/langfuse.jsonl.gz

Where the log lands

In your Langfuse project, cloud or self-hosted.

How it reaches the harness

You download it with your own Langfuse keys; the harness scripts never call the network.

Prompt names and versions recorded in Langfuse become prompt versions in the console, so a prompt change that made an agent worse shows up automatically.

Read the full guide on GitHub

Langfuse

Import what you already have