Skip to main content

​
Overview

Langfuse captures agent traces. Kindred imports those traces so you can use Behavior Search, Clustering, Expectations, and Replay on real production behavior. This guide gets traces into Langfuse, connects Langfuse to Kindred, syncs them, and opens the first imported session. If your agent already uses Langfuse, skip to Step 2.

​
Step 1: Create a Langfuse project

  1. Create a Langfuse account at Langfuse Cloud.
  2. Create a project.
  3. Open project settings.
  4. Copy the public key, secret key, and correct host.
Kindred supports these Langfuse Cloud hosts:

​
Step 2: Add Langfuse tracing

Use Set Up with a Coding Agent if you want Cursor, Claude Code, Codex, or another coding agent to make the integration patch. The coding agent should:
  • Install the official Langfuse SDK or framework integration.
  • Configure the Langfuse credentials.
  • Trace the full agent workflow.
  • Preserve Kindred metadata such as agent_id, session_id, and run_id.
  • Verify that a trace appears in Langfuse.
Your agent should have these values in its .env file:

​
Step 3: Verify the trace in Langfuse

Run one real agent request, then open the trace in Langfuse. Confirm the trace includes:
  • Input
  • LLM and agent operations
  • Tool calls and results
  • Final output

​
Step 4: Connect Langfuse to Kindred

​
Register a Kindred agent - app.usekindred.dev

Create a Kindred agent so imported traces have an agent to attach to.
  1. In Kindred, open Register agent.
  2. In New agent, enter an Agent name.
  3. Under Trace source, choose Langfuse.
  4. Click Create.
  5. Copy Agent ID (use as KINDRED_AGENT_ID in tracer).
  6. Paste that value into your agent .env file as KINDRED_AGENT_ID.

​
Connect Langfuse

  1. In Kindred, open Settings, then Langfuse.
  2. Click Create integration.
  3. Choose the agent, or leave it as Any agent if traces include agent_id.
  4. Click Configure.
  5. Paste the Langfuse public key, secret key, and host.
  6. Keep Enable scheduled pull sync on.
  7. Click Save and validate.
  8. Click Sync now.
Use the same Langfuse values in your agent .env file. See Environment Variables for the exact names to set. Kindred validates the host and API keys when you save. If validation cannot find a trace, run one traced agent turn in Langfuse, then save again. Confirm the trace appears in Behavior Search.

​
Step 5: Open the trace in Kindred

  1. Open Behavior Search.
  2. Select the agent.
  3. Open the imported session.
  4. Inspect its turns and steps.

​
Step 6: Add a replay URL

Replay requires an endpoint Kindred can call to rerun the selected agent turn. If your agent is deployed, use its replay endpoint, for example:
If your agent only runs locally, expose it with a tunnel and paste the public tunnel URL into Kindred.
Then use the public HTTPS URL as Target URL (replay_url). You can use any public HTTPS endpoint, such as a deployed service, ngrok, Cloudflare Tunnel, or similar. Use Configure Replay URL if your agent does not have a replay endpoint yet. That page includes a coding-agent prompt that can add one for you. You can also use Prompt 2 in Set Up with a Coding Agent.

​
Step 7: Run replay

In Kindred, open Behavior Search, click the trace or turn you want to replay, click Replay turn, paste the replay URL from Step 6, and click Run. Kindred will call your replay URL, find the replay trace in Langfuse, and show the first important difference it finds.

​
Core terms

  • Session: a conversation or workflow.
  • Turn: one logical agent invocation.
  • Step: an LLM call, tool call, tool result, or operation.