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

# Ragas

> Evaluate LLM apps with Ragas and AI Sonar

## Overview

Ragas can evaluate AI Sonar-backed applications by passing an OpenAI-compatible `AsyncOpenAI` client into `llm_factory`.

<Note>
  **Type**: Evaluation framework

  **Primary Path**: OpenAI-compatible Chat Completions

  **Support Confidence**: Supported OpenAI-compatible path
</Note>

## Environment

```bash theme={null}
export API_KEY="sk-your-api-key"
```

## Example Evaluator

```python theme={null}
import os

from openai import AsyncOpenAI
from ragas.llms import llm_factory
from ragas.metrics import DiscreteMetric

client = AsyncOpenAI(
    api_key=os.environ["API_KEY"],
    base_url="https://api.aisonar.dev/v1",
)
llm = llm_factory("claude-sonnet-5", client=client)

metric = DiscreteMetric(
    name="summary_accuracy",
    allowed_values=["accurate", "inaccurate"],
    prompt="Evaluate whether the response is accurate. Answer only accurate or inaccurate.\n\nResponse: {response}",
)
```

Use `llm` in Ragas metrics and testsets the same way you would use an OpenAI SDK-backed model.

## Endpoint Notes

Ragas uses the OpenAI SDK client path here. Native AI Sonar Responses, Anthropic Messages, and Gemini routes are best used through evaluation runners that support those request shapes directly.