Skip to main content
Connect Langfuse to inspect conversation activity from agents in your selected workspace. You can use Langfuse alongside Datadog; each connection has its own credentials and export behavior.

​
Before you begin

You need Admin or Owner access to the workspace and a Langfuse project. Create a public and secret key pair in your Langfuse project’s Settings → API Keys, and use the host for that project’s region. See Langfuse’s setup guide for project credentials.

​
Connect your project

  1. Select the intended workspace, then open Resources → External integrations → Langfuse.
  2. Enter the Host, Public Key, and Secret Key from the same Langfuse project. Enter the base host, such as https://cloud.langfuse.com, without an API endpoint path.
  3. Select Save.
  4. Run a new browser test or inbound call, then open your Langfuse project and check for its trace.
Saving credentials does not verify trace delivery. If a trace is missing, confirm that the host matches the project’s region, the key pair belongs to that project, and you are viewing the correct time range. Run a new conversation after correcting the connection.

​
Review traces

Inspect the recorded model activity, tool calls, timings, and errors to understand how the conversation progressed. Available details depend on the services used and the tracing configuration. See review traces for what to look for, and compare the trace with the run record.
Langfuse traces may contain prompts, conversation content, and tool data. Grant access to the project only to people who need to review that content. Datadog’s content exclusions do not apply to the separate Langfuse connection.

​
Update or remove the connection

Saved keys are masked. To rotate credentials, enter the new project’s key pair and select Save. Select Remove to stop using this workspace’s saved Langfuse connection. Removing it does not delete traces already stored in Langfuse.