# RAGView

RAGView helps you evaluate and select RAG solutions using real business data. Without complex deployment, you can compare retrieval quality, answer quality, latency, and resource consumption across different RAG strategies and choose the approach that best fits your use case.

RAGView provides two core modules:

- **RAG Space**: Browse RAG solutions integrated with the platform, view their details, and start evaluations for supported solutions.
- **RAG Arena**: Upload a document, add multiple RAG solutions, and compare their answers and runtime metrics on the same question.

> Evaluations consume model and application sandbox resources and are billed according to actual usage.

## Supported Scope

- Online evaluation is currently available for solutions marked **Evaluable** on the RAG Space page. Available solutions and versions are subject to the console.
- Supported document formats are PDF, DOCX, TXT, MD, and XLSX. Each file must not exceed 10 MB.

## Prerequisites

Before using RAGView, make sure that:

1. Your project has a valid API key with sandbox access and permission to call the required text-generation and embedding models.
2. Your account has sufficient balance to cover model and sandbox resource usage.

## Quick Start

The complete evaluation workflow consists of four steps:

| Step | Action | Description |
| --- | --- | --- |
| 01 | Upload a document | Use a document as the knowledge-base source. |
| 02 | Prepare a test question | Enter a question, reference answer, and optional reference context. |
| 03 | Select RAG solutions | Choose the solutions to evaluate and compare. |
| 04 | Start the evaluation | Configure the models and evaluators, then submit the task. |

## Discover RAG Solutions

Go to **Model Service Platform > Experience Center > RAGView**. The **RAG Space** page is displayed by default.

Each solution card contains:

| Item | Description |
| --- | --- |
| Name and version | The solution identifier and latest available version. |
| Overview | Recommended use cases and technical characteristics. |
| Open-source status | Whether the solution is open source. Open-source solutions also show the license, GitHub source, and star count. |
| Evaluable status | Solutions marked **Evaluable** can be evaluated directly. |

## Evaluate Solutions in RAG Arena

Open the **RAG Arena** tab to compare multiple RAG strategies on the same question.

### 1. Upload a Document

In the **Documents** area on the left, click **Select Document** and upload one document as the knowledge-base source.

- PDF, DOCX, TXT, MD, and XLSX files are supported, up to 10 MB per file.
- After uploading, click **Replace File** to select a different document.
- A snapshot is created when an evaluation task is submitted. Replacing the document later does not affect submitted tasks.

> The document is sent to a cloud sandbox to build the knowledge base. Do not upload files containing plaintext credentials or sensitive personal information.

### 2. Configure Evaluators (Optional)

In the **Evaluators** area, select the metrics you want to calculate. RAGView currently supports platform-provided evaluators. Their definitions and calculation logic are available on the **Evaluators** page.

Some metrics require an LLM, an embedding model, a reference answer, or reference context.

### 3. Add RAG Solutions

In the comparison area, click **Add Solution** and select up to four evaluable RAG solutions. Each solution appears on a separate card and can be removed independently.

### 4. Ask a Question and Compare Results

Enter a test question at the bottom of the page. All selected solutions run retrieval and question answering in parallel. Model and sandbox usage is billed according to actual consumption.

During execution, each card shows its current status and stage logs. When execution completes, the card displays:

| Item | Description |
| --- | --- |
| Generated answer | The complete Markdown-formatted answer. |
| TTFT | Time from request submission to the first output token. |
| Total duration | Total execution time for the request. |
| Token usage | Total tokens consumed by the request. |
| Retrieved context | Number of retrieved context chunks; expand to view details. |
| Evaluator scores | Scores produced by the selected evaluators. |
| Error details | Failure reason, if any. One failed solution does not affect the others. |

## View Evaluators

Open the **Evaluators** tab to browse all platform-provided metrics. Click **View Details** on a metric card to inspect:

| Item | Description |
| --- | --- |
| Category | Metric type: LLM or Code. |
| Objective | Metric identifier and evaluation objective. |
| Source | Origin of the metric. |
| Input | Required fields and formats, such as `question`, `contexts`, and `reference_answer`. |
| Output | Output type and value range. |
| Function logic | The implementation or LLM prompt, which can be viewed and copied. |

## Billing

- RAGView itself has no feature fee. You are billed only for model calls and application sandbox resources consumed during evaluation.
- Token usage and sandbox runtime are subject to the billing records. RAGView usage details are available under **Billing Center > Export Records**.
- LLM-based evaluators generate additional token usage. Select evaluators according to your accuracy requirements.

## FAQ

### Why Does a Solution Show “Failed” or “inference failed”?

Common causes include insufficient model permissions, insufficient balance, document parsing failure, or sandbox resource errors. Verify the API key permissions, account balance, and document contents, then retry. A failed solution does not affect other solutions and can be removed and added again.

### Can I Export Evaluation Results?

Currently, complete answers, retrieved context, and metric scores can only be viewed in the console. Saving and exporting evaluation tasks and results will be supported in a future release.