Retrieval

R2R’s Retrieval system provides advanced search and generation capabilities powered by vector search, knowledge graphs, and large language models. The system offers multiple ways to interact with your data:

• Direct semantic search across documents and chunks
• Retrieval-Augmented Generation (RAG) for AI-powered answers
• Conversational RAG agents for complex queries
• Raw LLM completions for flexible text generation

Core Features

Vector Search

• Semantic similarity matching using document/chunk embeddings
• Hybrid search combining vector and keyword approaches
• Complex filtering with Postgres-style operators
• Configurable search limits and thresholds

Knowledge Graph Search

• Entity and relationship-based retrieval
• Multi-hop traversal for connected information
• Local and global search strategies
• Community-aware knowledge structure

RAG Generation

• Context-aware responses using retrieved content
• Customizable generation parameters
• Source attribution and citations
• Streaming support for real-time responses

RAG Agent

• Multi-turn conversational capabilities
• Complex query decomposition
• Context maintenance across interactions
• Branch management for conversation trees

Available Endpoints

Search Settings

Vector Search Settings

1
{
2
  "use_semantic_search": true,
3
  "filters": {"document_id": {"$eq": "3e157b3a-8469-51db-90d9-52e7d896b49b"}},
4
  "limit": 20,
5
  "use_hybrid_search": true
6
}

Knowledge Graph Settings

1
{
2
  "enabled": true,
3
  "generation_config": {
4
    "model": "gpt-4o-mini",
5
    "temperature": 0.7
6
  }
7
}

Generation Configuration

1
{
2
  "stream": false,
3
  "temperature": 0.7,
4
  "max_tokens": 150,
5
  "model": "gpt-4o-mini"
6
}

Key Concepts

Search

The search endpoint provides direct access to R2R’s retrieval capabilities, allowing you to find relevant content using semantic similarity and knowledge graph relationships. Results can be filtered using complex queries and sorted by relevance.

RAG

RAG combines retrieval with language model generation to produce informative responses grounded in your content. The system retrieves relevant context and uses it to generate accurate, sourced answers to queries.

Agent

The RAG agent provides a conversational interface for complex information retrieval. It can maintain context across multiple interactions, break down complex queries, and provide detailed responses with citations to source material.

Completion

Direct access to language model generation capabilities, useful for tasks that don’t require retrieval from your content. Supports both single-turn and multi-turn conversations.

Filter Operations

Supported operators for content filtering:

• eq: Equals
• neq: Not equals
• gt: Greater than
• gte: Greater than or equal
• lt: Less than
• lte: Less than or equal
• like: Pattern matching
• ilike: Case-insensitive pattern matching
• in: In list
• nin: Not in list

Example:

1
{
2
  "filters": {
3
    "metadata.category": {"$eq": "research"},
4
    "created_at": {"$gte": "2024-01-01"},
5
    "collection_ids": {"$in": ["uuid1", "uuid2"]}
6
  }
7
}

Common Use Cases

1. Research and Analysis

   • Literature review
   • Document summarization
   • Relationship discovery
   • Cross-reference verification

2. Question Answering

   • Technical support
   • Educational assistance
   • Policy compliance
   • Data exploration

3. Content Generation

   • Report writing
   • Documentation creation
   • Content summarization
   • Knowledge synthesis

4. Conversational Applications

   • Interactive chatbots
   • Virtual assistants
   • Educational tutors
   • Research aids