Skip to main content
Use this path when you want the fastest Kindred setup and do not want to wire trace code by hand. Your coding agent will use Kindred’s public skills repo to register your agent, install kindred-tracing, instrument the real turn path, verify traces, and add replay when possible. Use Langfuse if your team already has Langfuse tracing in production. Use OTel when you want a direct Kindred integration that a coding agent can add quickly.

​
Overview

You will:
  1. Create a Kindred account.
  2. Create a KINDRED_API_KEY.
  3. Give your coding agent the Kindred skills prompt.
  4. Let the coding agent register and instrument your agent.
  5. Open Kindred to inspect traces, replay, and other features.

​
Step 1: Create a Kindred account

Sign in to Kindred or create an account in the Kindred web app.

​
Step 2: Create an API key

In Kindred, open Settings → API Management and create or copy your API key. Make the key available in the shell where your coding agent can use it:
The coding agent can use this key with npx kindred-cli. You do not need to install the CLI manually.

​
Step 3: Open your agent repo

Open the codebase that contains your AI agent in Codex, Cursor, Claude Code, or another coding agent. Then paste this prompt:

​
Step 4: Let the coding agent work

The coding agent should do the rest:
  • Use npx kindred-cli or an installed kindred binary.
  • Register your agent with --trace-source otel.
  • Install kindred-tracing.
  • Add KINDRED_AGENT_ID and KINDRED_API_KEY to env docs.
  • Find the real agent turn entrypoint.
  • Wrap one user turn in one root OpenTelemetry trace.
  • Record LLM calls and tool calls as child spans.
  • Run one real prompt.
  • Verify the trace appears in Kindred.
  • Add POST /replay if your app has an HTTP server.

​
Step 5: Open Kindred

When the coding agent finishes, it should give you:
  • Agent ID
  • Session ID
  • Trace ID
  • Replay result, if replay was configured
  • Product links
Click the links the coding agent provides, or open Kindred and select the registered agent. Start with Behavior Search to inspect the traced session. From there, you can:
  • Use Replay to rerun a turn and compare behavior.
  • Use Expectations to define behavior rules.
  • Use Clustering to find recurring patterns once the agent has enough traces.

​
Quick check

Before you consider setup complete, confirm:
  • The agent is registered with trace source otel.
  • A real prompt created a Kindred session.
  • The session has at least one turn.
  • The turn includes LLM generation spans, and tool spans if your agent uses tools.
  • Replay is configured if your agent has a public /replay endpoint.