> ## Documentation Index
> Fetch the complete documentation index at: https://whyops.com/docs/llms.txt
> Use this file to discover all available pages before exploring further.

# Introduction

> WhyOps: Observability and Evaluations for AI Agents

**WhyOps is a decision infrastructure platform for autonomous agent systems.**

We move beyond basic prompt and token logging to help you understand exactly why your agents make the decisions they do. WhyOps runs in a proxy architecture that sits seamlessly between your application and your model providers.

<Callout type="success" title="Deployed and Production Ready">
  WhyOps is currently deployed in production. You can proxy requests, evaluate agent runs, and view decision graphs in real-time.
</Callout>

## The Problem

Traditional observability tools fail when it comes to autonomous AI agents because they track deterministic execution paths (Request → Response). Agents, however, operate using:

* Non-deterministic planning loops
* Internal chain-of-thought and invisible reasoning
* Tool calling loops and failure retries
* Dynamic context assembly

When an agent fails, you need to answer:

* *Why did it invoke that specific tool?*
* *Why did it enter an infinite retry loop?*
* *How did the system prompt influence this specific chain of thought?*

## The WhyOps Solution

WhyOps intercepts LLM traffic to trace these internal decision loops automatically. We do this primarily through our **Proxy Layer** that captures:

1. **Complete LLM Interactions**: Prompts, tool calls, and LLM responses.
2. **Invisible Signatures**: A mechanism that injects hidden tokens to trace context across disjointed API requests, stitching them into a unified agent "Thread."
3. **Agent Evaluations**: Our `whyops-analyse` service runs evaluations across traces, scoring runs on dimensions like intent precision, tool routing quality, and reliability.

## Key Features

<CardGroup cols={2}>
  <Card title="Official SDK Packages" icon="box-open" href="/docs/integrations/sdk-packages">
    Use `@whyops/sdk`, `whyops`, or the Go module to combine agent init, proxy patching, and manual runtime events in one integration surface.
  </Card>

  <Card title="Invisible Signatures" icon="fingerprint" href="/docs/core-concepts/invisible-signatures">
    Automatically groups stateless LLM calls into coherent traces without requiring complex client-side SDK instrumentation.
  </Card>

  <Card title="Automated Static Analysis" icon="chart-line" href="/docs/core-concepts/evaluations-and-analysis">
    Detects duplicate tool calls, missing responses, missing context, and infinite retry loops passively.
  </Card>

  <Card title="Proxy Architecture" icon="network-wired" href="/docs/architecture/proxy-service">
    Drop-in replacement for OpenAI and Anthropic API endpoints. Just change your Base URL.
  </Card>

  <Card title="Decision Graph UI" icon="diagram-project" href="/docs/architecture/frontend-app">
    Visualize the precise causal chain of reasoning steps that led to an agent's final answer.
  </Card>
</CardGroup>