The observability stack for AI agents.

Trace your AI agents with a visual workflow pipeline. Rackle captures prompts, tool calls, contexts, and errors in a clean interface that makes debugging complex LLM systems intuitive.

Star on Github
Support Access Workflow
Prompt

You are the Support Operations Agent...

Context

Unexpected charge for external agency...

Tool
add_member0.6s
Input:{"email":"agency@partner.co"}
Output:{"status":"member created"}
Model

Chose add_member to give the agency access as a viewer. Result: the user was added as a billable member.

Error

Guardrail failed: Non-employee domain detected for billable role.

Everything you need to debug AI

Rackle provides a complete suite of tools to monitor, analyze, and improve your AI agents in production.

Real-time Tracing

Watch your agents execute step-by-step. See prompts, tool calls, and latency instantly as they happen.

AI Copilot

Ask Rackle AI to analyze traces, find bottlenecks, and explain errors using your actual execution data.

Cost & Tokens

Track token usage across all your models. Get detailed analytics on your spend and optimize your pipelines.

How It Works

From code to dashboard in three simple steps.

1

Install the SDK

Drop the lightweight Rackle SDK into your Node.js or Bun project. It's fully typed and has zero external dependencies.

2

Log your Steps

Wrap your LLM calls, tool executions, and memory operations with run.log(). Rackle handles the async batching automatically.

3

Analyze on the Dashboard

View beautiful, nested execution traces. Identify failures instantly, track latency, and use Rackle AI to debug your agents.

Integrate in two lines of code.

Our lightweight TypeScript SDK drops perfectly into any Node.js, Bun, or Edge environment. Capture rich step-by-step executions automatically.

Type-safe SDK Zero dependencies
agent.ts
import { Tracer } from "@rackle-labs/sdk";

const tracer = new Tracer({
apiKey: process.env.RACKLE_API_KEY
});

const run = await tracer.startRun({ agentName: "Customer-Bot" });

await run.log({
type: "llm_call",
model: "gpt-4o",
tokens: 350,
input: "Help me reset my password."
});

await run.end({ status: "completed" });

Frequently Asked Questions

Rackle is entirely model-agnostic. You can log traces from OpenAI, Anthropic, local open-source models, or any other LLM provider. Simply pass the model name and token usage into the log step.

Yes. Rackle is designed to run locally or be self-hosted in your own infrastructure. Your traces, prompts, and API keys never leave your environment.

No. The SDK batches logs asynchronously and sends them in the background, ensuring your critical AI paths remain extremely fast and uninterrupted.

Stop Guessing.
Start Monitoring.

Get total observability into your AI agents in minutes. No complex setups, no agent overhead, just clean trace outputs.