Tensile Docs
Integrations

Langfuse

Connect a Langfuse project so Tensile can import selected traces for workflow discovery and review.

Watch the workflow: Prepare a Langfuse connection draft · Authored dummy data

This recording uses the earlier Connections layout. In the current app, open Settings → Integrations → Langfuse. Select Connect to open a new connection form, or the pencil Edit connection icon beside the saved connector's name and status. The host, credential, customer-key and backfill draft fields remain available. The recording does not save a connection or show the current logs view.

Fill a local demo host, dummy provider keys, tenant field and backfill date without connecting or running an external sync.

Read matching command examples.

Read the walkthrough transcript
  • 00:00 — Prepare a Langfuse host in Tensile. This local demo stays unsaved.
  • 00:05 — Enter dummy keys. A real source needs your authorized provider credentials.
  • 00:11 — Choose tenant identity and a backfill date. This Langfuse draft stays unconnected.

Connect an existing Langfuse project when your agent traces are already stored there. Tensile reads traces and observations through the Langfuse API and maps them into sessions and events for the workspace. It does not require you to re-instrument the same agent just to import those records.

Connect a project

  1. In Langfuse, create an API key pair for the project. Copy the public key, secret key, and the host for your Langfuse instance.
  2. In Tensile, open Settings → Integrations → Langfuse and select Connect. An organization owner or admin can manage connections.
  3. For an existing connection, select the pencil Edit connection icon, then enter the Host, Public key, and Secret key. For an existing connection, leave the secret field blank to keep the saved key.
  4. Optionally set the Customer-id metadata key (default customer_id) and Backfill from date. Leave the date blank to request all available history.
  5. Select Connect or Update. Connected means the configuration is saved. The current form does not show synchronization progress or verify provider access.

The secret key is used by the server-side connector. Keep it out of source control, browser code, and support messages. Use credentials with only the access needed to read the selected project.

View logs for the connected project

Select View logs on a saved Langfuse connection to open /settings/integrations/logs?provider=langfuse. The connection's keys identify the project; this page does not offer a separate project selector. Select a row to read its full-width request detail, then use the breadcrumb to return to the scoped list. A saved connection or an empty list does not prove current provider access or complete trace import. The authenticated /raw-logs reader and direct SDK, Pulse and Error Captures links remain available even though Raw Logs is not in main navigation.

To disconnect the saved connector, select the destructive Disconnect icon beside Edit connection in the header. The confirmation dialog names the action; Cancel keeps the connection.

Check project access from a terminal

Load LANGFUSE_HOST, LANGFUSE_PUBLIC_KEY, and LANGFUSE_SECRET_KEY from your approved secret store. Use the host for your Langfuse project, including its region or self-hosted address. Use the same host in Tensile.

curl --fail --silent --show-error \
  --user "${LANGFUSE_PUBLIC_KEY}:${LANGFUSE_SECRET_KEY}" \
  "${LANGFUSE_HOST}/api/public/projects"

A successful response lists the project associated with the key pair. This checks provider access. It does not save a Tensile connection or prove that traces were imported. See the Langfuse public API reference.

What is imported

Tensile maps Langfuse traces into sessions and observations into events. Trace and observation identifiers are retained as source references where available; the configured customer-id metadata key can be used to associate a downstream customer. Exact fields depend on the data present in the Langfuse project.

Imported traces can support workflow discovery and evidence review. Connecting Langfuse does not approve a pack or activate evaluator-pack scoring. See Evaluator packs for the review and validation gates.

On this page