Skip to main content

​
Overview

Kindred lets you capture, search, and replay your agents’ behavior with minimal code changes. Use the Tracer SDK to auto-instrument your calls and the Replay Studio to debug sessions in real-time.

Tracer SDK

Auto-instruments LLM + tool calls via Node/Python.

Replay Studio

Debug and analyze sessions visually.

Reproducibility

Deep-link into specific sessions and drift checks.

​
1. Create an Agent in Kindred

Follow these steps to register your agent and obtain your credentials:
  1. Sign In: Log into the Kindred web app.
  2. Register Agent: Go to Agents → New Agent and provide a Name and your Run Step URL (the HTTPS endpoint where Kindred can call your agent).
  3. Save Credentials: Copy the Bearer token, Agent ID, and API key.
The Bearer token and API key are only shown once! Save them to a secure location immediately.

​
2. Environment Variables

Set these in your agent’s environment (e.g., .env file or CI/CD secrets).

​
Tracer Configuration

VariableDescription
KINDRED_API_KEYSecret token allowing the tracer to send logs.
KINDRED_AGENT_IDUnique ID used to attribute logs to this agent.
KINDRED_SESSION_ID(Optional) Logical ID for a conversation or user session.
KINDRED_RUN_ID(Optional) Identifier for a specific experiment or AB test.

​
Server Validation

  • KINDRED_BEARER_TOKEN: Used by your agent’s HTTP server to verify that incoming requests are actually from Kindred.

​
3. Install and Initialize the Tracer

​
Node.js (kindred-tracer-node)

Install the package using your preferred manager: 
npm install kindred-tracer-node
Initialize the tracer at the very top of your entry point: 
import { kindredTracer, flushLogs } from "kindred-tracer-node";

async function main() {
  // Initialize once at startup (reads from process.env)
  kindredTracer();

  // Standard LLM call - auto-captured by tracer
  const res = await fetch("[https://api.openai.com/v1/chat/completions](https://api.openai.com/v1/chat/completions)", {
    method: "POST",
    headers: {
      "Content-Type": "application/json",
      Authorization: `Bearer ${process.env.OPENAI_API_KEY}`,
    },
    body: JSON.stringify({
      model: "gpt-4o-mini",
      messages: [{ role: "user", content: "Hello Kindred!" }],
    }),
  });

  // Ensure logs are sent before process exit
  await flushLogs();
}

​
Python (kindred-tracer)

Install the package with HTTP support: 
pip install kindred-tracer[all]
Initialize the tracer before creating your LLM client: 
from kindred_tracer import kindred_tracer, flush
from openai import OpenAI
import os

# Initialize tracer (reads KINDRED_* env vars)
kindred_tracer()

client = OpenAI(api_key=os.environ["OPENAI_API_KEY"])
response = client.chat.completions.create(
    model="gpt-4o-mini",
    messages=[{"role": "user", "content": "Hello Kindred!"}],
)

# Flush logs before shutdown
flush()

​
4. Connecting Your Run Step URL

Kindred calls your agent’s HTTP endpoint to enable validation and replay.

​
Validation Request

Kindred will send a POST request to your configured URL with an Authorization: Bearer <KINDRED_BEARER_TOKEN> header. Request Body: 
{
  "observation": {
    "type": "kindred_validation",
    "message": "hello"
  }
}
Local Development: Use a tunnel like ngrok to expose your local server via HTTPS so Kindred can reach your machine.

​
5. Linking to Replay URLs

You can programmatically generate deep links to specific sessions to include in your internal dashboards, Sentry issues, or Slack alerts. URL Format: https://app.usekindred.dev/reproducibility?sessionId=<ID>&agentId=<ID> 
function getReplayUrl(sessionId, agentId) {
  const baseUrl = "[https://app.usekindred.dev/reproducibility](https://app.usekindred.dev/reproducibility)";
  return `${baseUrl}?sessionId=${sessionId}&agentId=${agentId}`;
}

​
6. Quick Checklist

  • Created agent in Kindred and saved all tokens.
  • Installed the Tracer SDK for your language.
  • Environment variables are set in your production/local environment.
  • Called initialization function at the start of your app.
  • Run Step URL is validated and returns {"ok": true}.
Once complete, your agent is fully wired for search, replay, and reproducibility!