Skip to content

RAG Agent

The RAG (Retrieval-Augmented Generation) Agent provides a natural language interface for exploring and searching document collections. It uses RAG techniques to combine search capabilities with generative AI to produce relevant, context-aware responses based on document content.

Features

  • Search document collections with natural language queries
  • Inspect specific documents by title or other identifying information
  • Retrieve PubMed publications via PMID
  • Perform web searches and retrieve web page content
  • Generate context-aware responses based on document content

Usage

Python API

from aurelian.agents.rag import rag_agent, get_config

# Initialize dependencies with your database path and collection
deps = get_config(db_path="sqlite:///path/to/db", collection_name="main")

# Search for documents about a specific topic
result = await rag_agent.run(
    "What papers in collection are relevant to microbial nitrogen fixation?",
    deps=deps
)

# Get the search results
print(result.data)

Command Line Interface

aurelian rag --db-path "sqlite:///path/to/db" --collection-name "main" "What papers in collection are relevant to microbial nitrogen fixation?"

Gradio Interface

from aurelian.agents.rag import chat

# Launch Gradio interface with your database configuration
interface = chat(db_path="sqlite:///path/to/db", collection_name="main")
interface.launch()

Tools

The RAG Agent provides the following tools:

search_documents

Performs a retrieval search over the document database using a natural language query. Returns matching documents with relevancy scores.

inspect_document

Retrieves the full content of a specific document identified by title or other identifying information.

lookup_pmid

Retrieves the full text or abstract of a publication using its PubMed ID (format: "PMID:nnnnnnn").

search_web

Searches the web using a text query and returns matching web pages with summaries.

retrieve_web_page

Fetches the contents of a specific web page.

Configuration

The RAG Agent requires a document database to function. You can configure it through:

  • Environment variables:
  • AURELIAN_RAG_DB_PATH: Path to the document database

  • AURELIAN_RAG_COLLECTION: Name of the collection (default: "main")

  • Direct configuration:

  • Pass db_path and collection_name parameters to get_config()

Examples

Example 1: Searching for papers on a specific topic

What papers in collection are relevant to microbial nitrogen fixation?

Example 2: Finding papers that mention specific genes

Find documents that discuss the role of nif genes in nitrogen fixation

Example 3: Summarizing a specific document

Summarize the paper titled "Recent advances in nitrogen fixation"