AutoRAG
AutoRAG enables intelligent search and retrieval from the knowledge graphs built by the Importer and AutoGraph
Overview
AutoRAG provides intelligent search and retrieval from the knowledge graphs built by the Importer and AutoGraph. Its search methods rely on the specific structure these services create, with entities, relations, and community summaries, so they cannot run against an arbitrary graph of your own. AutoRAG offers multiple search methods optimized for different query types, from fast instant answers to thorough multi-step research, plus a Custom Retriever that can additionally search any collection in the database using search logic you define yourself.
graphragretriever deployment
endpoint and in the /graphrag/retriever/ URL prefix of the query API, and in
AutoGraph Studio,
which calls the services it deploys retrievers.When to use AutoRAG
AutoRAG serves two usage patterns depending on how your knowledge graph was built.
With the Importer (standalone)
When using the Importer directly to build a single knowledge graph, you query it with the query API, which gives access to all search methods and parameters.
The standalone Importer workflow is not available in the web interface. AutoGraph Studio can only query Context Graphs that belong to an AutoGraph project.
With AutoGraph (partitioned)
When AutoGraph manages your document pipeline, it builds
partitioned knowledge graphs with domain-aware RAG strategies. AutoRAG queries
across these partitions using partition_ids to target specific domains. AutoGraph’s two-stage retrieval pattern first identifies relevant
partitions, then performs deep search within them.
Each partition is built either as VectorRAG or as FullGraphRAG, and that choice determines which search methods you can run against it. See VectorRAG and FullGraphRAG partitions.
For details on how partitions are created and mapped, see the Importer AutoGraph Integration page.
Search methods
| Method | Best for | Latency |
|---|---|---|
| Unified (Instant Search) | Fast answers with document references | Low |
| Deep Search | Thorough, multi-step research | Higher |
| Global Search | Themes, patterns, high-level insights | Medium |
| Local Search | Specific entities and relationships | Low |
| Custom Retriever | Domain-specific search on any collection | Varies |
See Search Methods for a full comparison and guidance on choosing the right method.
Prerequisites
Before using AutoRAG, you need:
A project with imported data. If you are using the Importer standalone, follow the Importer Setup. If you are using AutoGraph, follow the AutoGraph Setup.
An LLM provider configured for AutoRAG. See LLM Configuration to set up Triton Inference Server or any OpenAI-compatible API.
Installation
To install and start AutoRAG, use the following endpoint:
For detailed instructions on installing, monitoring, and managing services, see The Arango Control Plane (ACP) service.
Getting Started
- Configure your LLM provider: Choose and configure either Triton or OpenAI-compatible APIs.
- Understand search methods: Learn about Instant, Deep, Global, Local, and Custom search.
- Execute queries: Start querying your knowledge graph.
Additional resources:
- Custom Prompts: Customize LLM prompts for domain-specific behavior.
- Verify and Monitor: Check service health, service status, and the query history of your project.
- Parameter Reference: Complete list of query parameters.
- Error Handling: Status codes, error codes, and the
failures that arrive with an HTTP
200.
API Reference
For detailed API documentation, see the GraphRAG Retrievers API Reference .