RAG Agent is a retrieval-augmented generation (RAG) solution that simplifies building production-grade RAG applications. It streamlines the entire data pipeline, from ingestion and processing to retrieval. Deeply integrated with the Supabase ecosystem, RAG Agent automatically processes documents from cloud storage and supports advanced features such as multimodal parsing and knowledge graph construction. This lets you efficiently add powerful, knowledge base-driven Q&A capabilities to your applications.
Features
Deep integration with Supabase
You can enable RAG Agent when you create an RDS Supabase project and manage table schemas and data through the Supabase visual interface. Additionally, it directly uses Supabase Auth, which simplifies managing user access permissions to datasets.Automated data processing and embedding
RAG Agent provides an end-to-end data processing pipeline. When you create a dataset and configure Supabase Storage as its storage backend, the agent automatically scans and syncs file updates. It then manages the entire workflow, including smart chunking, vector embedding, and knowledge graph construction, which greatly simplifies the data ingestion process.Multimodal document parsing
RAG Agent parses complex multimodal documents containing text, images, tables, and formulas. The system uses dedicated parsing models and smart chunking strategies to ensure elements such as images are correctly cited in Q&A results. The parsed intermediate files (middle.json) are also stored locally to support custom development or debugging.GraphRAG and multi-mode queries
RAG Agent supports building a knowledge graph which enhances retrieval accuracy and depth. You can choose from several query modes, including vector search, graph search, or a hybrid search that combines both, to suit different business scenarios. The system's query performance is also optimized for PostgreSQL.
Core features
Dataset Management: Create, query, update, and delete datasets, and view dataset statistics and health checks.
Document Management: Provides a unified document upload interface that supports local files and cloud storage services such as Supabase and Alibaba Cloud OSS. You can list, delete, clear, and scan documents for updates.
RAG Query: Query single or multiple datasets with streaming responses, using one of several built-in query modes.
Query mode
Use case
Response speed
Accuracy
Resource consumption
naive
Vector search.
Fastest
Medium
Lowest
local
Performs retrieval based on local context. Ideal for specific questions that require precise answers.
Fast
High
Low
global
Queries the global knowledge graph. Suitable for general or conceptual questions that require a deep understanding.
Medium
High
Medium
hybrid
Combines local and global knowledge for a hybrid search that balances accuracy and coverage.
Medium
High
Medium
mix
Combines the naive and hybrid modes, performing both vector search and graph search.
Medium
High
High
bypass
Bypasses RAG and sends the query directly to the large language model (LLM) for a response.
Fast
Depends on the LLM
Medium
Knowledge Graph: Retrieve graph data and labels for a dataset, and use the front-end interface to visualize, analyze, search, and navigate it.
Multimodal Processing: Integrates a professional document parsing service with advanced vision models (such as Qwen-VL and GPT-4 Vision) to process complex multimodal documents containing text, images, tables, and formulas. It supports multiple languages and optical character recognition (OCR).
Cache Management: Provides precise, dataset-level cache management.
Pipeline Management: Provides real-time status updates on document processing and supports pipeline monitoring through the front-end interface.
Storage Management: Supports multiple storage solutions to fit your project's cost, performance, and deployment requirements.
Storage type
Use case
Configuration complexity
Performance
Cost
Scalability
Local storage
Development, testing, and small-scale deployments
Low
Fast read/write speeds
Hardware costs
Limited
Supabase
Enterprise production environments
Low
Integrates with RDS PostgreSQL and Alibaba Cloud OSS
Cost-effective
Very high
Alibaba Cloud OSS
Enterprise production environments
Medium
High availability and global acceleration
Pay-as-you-go
Very high
Amazon S3-compatible object storage
Large enterprises and global deployments
Medium
Mature object storage solution
Multiple billing models
Very high
Data isolation and security: Isolates data for each dataset using PostgreSQL schemas, providing enterprise-grade data isolation and security.
Monitoring and Diagnostics: Provides a comprehensive health check API, pipeline status monitoring, and robust error handling and feedback.
Front-end interface: Provides a front-end interface for dataset management, document management, querying, knowledge graph visualization, and interactive API documentation. It supports multiple languages and local configurations.
Enable and use RAG Agent
When you create an RDS Supabase project, enable Enable RAG.
RAG Agent includes built-in models. If you use external custom models that require public network access, you must enable Internet NAT Gateway: for the agent.
After the project is created, in the RDS Supabase list, click Is set White List in the Access address column for your project.
On the Basic Information page, go to the White list information section, click Create Whitelist, and add your client's IP address to the allowlist.
In your browser, enter
http://<Supabase public access address>/ragto open the RAG Agent login page.ImportantThe Supabase public access address must include the port number. For example:
http://8.147.XXX.XXX:80/rag.If you are logging in from an ECS instance within the same VPC as your RDS Supabase project, we recommend using the internal access address.
Enter the dashboard user (the default is supabase) and your RDS Supabase project password to log in.