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

# Introduction

> Knowledge reasoning engine for enterprise AI. Traceable answers from structured knowledge graphs.

Vrin is a **knowledge reasoning engine** that structures your documents into a temporal knowledge graph and reasons across it, giving AI answers you can trace to specific facts.

## Why Vrin

Traditional retrieval systems rely on text similarity alone, which breaks down when questions span multiple documents or require temporal context. Vrin adds a **knowledge graph layer** that:

* Extracts structured facts (entities, relationships, timestamps) from every document
* Traverses entity connections to answer questions that span dozens of documents
* Reasons across your knowledge graph to assemble precisely the context your LLM needs
* Traces every answer to specific facts from specific source documents

## Key features

<CardGroup cols={2}>
  <Card title="Deep Reasoning" icon="diagram-project">
    Knowledge graph reasoning that answers questions spanning dozens of documents. Not just keyword matching.
  </Card>

  <Card title="True streaming" icon="bolt">
    Server-Sent Events deliver tokens as they are generated. No buffering, no polling.
  </Card>

  <Card title="File upload" icon="file-arrow-up">
    Upload PDFs, CSVs, and text files. Vrin extracts facts and chunks automatically.
  </Card>

  <Card title="Enterprise data sovereignty" icon="shield-halved">
    Enterprise API keys route queries through your own AWS account. Your data never leaves your cloud.
  </Card>
</CardGroup>

## How it works

Every query follows a three-stage pipeline:

1. **Structure** -- VRIN extracts entities from your query and traverses the knowledge graph to find relevant facts, relationships, and temporal context.
2. **Reason** -- Graph facts and document chunks are intelligently fused, scored for confidence, and assembled into precisely the context your LLM needs.
3. **Trace** -- Every answer includes the specific facts used, the source documents they came from, and confidence scores, so you can verify every claim.

## For AI Agent Builders

VRIN works as a drop-in reasoning layer for AI agents. Integrate via Python SDK, MCP server, or REST API:

```python theme={null}
from vrin import VRINClient

client = VRINClient(api_key="vrin_live_your_api_key")
result = client.query("What changed in ACME's revenue between Q2 and Q3?")

# Every answer traces to specific facts
for source in result.get("sources", []):
    print(f"  {source['document']}: {source['fact']}")
```

VRIN is model-agnostic. It works with GPT, Claude, Gemini, or any LLM. You bring the model, VRIN provides the structured reasoning.

## Benchmarks

| Benchmark        |     VRIN     |             Best Baseline            | Improvement |
| ---------------- | :----------: | :----------------------------------: | :---------: |
| **MultiHop-RAG** |   **95.1%**  |     78.9% (GPT-5.2 w/ same docs)     |   +16.2pp   |
| **MuSiQue**      | **EM 0.478** | EM 0.372 (HippoRAG 2, academic SOTA) |     +28%    |

MultiHop-RAG: 384 stratified samples (seed=42), 609-article corpus. MuSiQue: 300 multi-hop questions, 4,848 paragraphs ingested.

## Next steps

<CardGroup cols={2}>
  <Card title="Quickstart" icon="rocket" href="/quickstart">
    Install the SDK and run your first query in under 2 minutes.
  </Card>

  <Card title="Python SDK" icon="python" href="/sdk/client">
    Full reference for VRINClient methods and configuration.
  </Card>

  <Card title="API Reference" icon="code" href="/api-reference/query">
    HTTP endpoints, request/response schemas, and interactive playground.
  </Card>

  <Card title="How it works" icon="lightbulb" href="/concepts/how-it-works">
    How the reasoning engine works: graph, facts, temporal versioning.
  </Card>
</CardGroup>