sam-rag 0.1.2

A document-ingesting agent that monitors specified directories, keeping stored documents up to date in a vector database for Retrieval-Augmented Generation (RAG) queries.

Solace Agent Mesh RAG

A document-ingesting agent that monitors specified directories, keeping stored documents up to date in a vector database for Retrieval-Augmented Generation (RAG) queries.

About Solace Agent Mesh

Solace Agent Mesh (SAM) is an open-source framework for building event-driven, multi-agent AI systems where specialized agents collaborate on complex tasks. It provides a standardized way for agents to communicate, share data, and integrate with external systems while keeping components loosely coupled and production-ready.

SAM helps you:

• Build event-driven multi-agent systems on Solace Event Mesh
• Connect agents, tools, gateways, and services through a common runtime
• Extend projects with installable plugins such as sam-rag

Learn more in the Solace Agent Mesh documentation and the main project repository.

Overview

The Solace Agent Mesh RAG system provides a complete RAG pipeline that includes:

1. Document Scanning: Monitors directories for new, modified, or deleted documents
2. Document Preprocessing: Cleans and normalizes text from various document formats
3. Text Splitting: Breaks documents into smaller chunks for embedding
4. Embedding Generation: Converts text chunks into vector embeddings
5. Vector Storage: Stores embeddings in a vector database for efficient retrieval
6. Retrieval: Finds relevant document chunks based on query similarity
7. Augmentation: Enhances retrieved content using LLMs

Documentation

Comprehensive documentation is available in the docs directory:

• Architecture Guide: Overview of the SAM RAG architecture and components
• Configuration Guide: Detailed explanation of configuration options
• Tools and Lifecycle Guide: Documentation for tools and lifecycle functions

Installation

Add the RAG Plugin to Solace Agent Mesh

solace-agent-mesh plugin add <your-new-component-name> --plugin sam-rag

This will create a new component configuration at configs/plugins/.yaml. You need to configure proper values by updating this file. Export at least the following environment variables to work with the default configuration. For more advance settings, please visit the Configuration Guide.

export SOLACE_BROKER_URL=ws://localhost:8008
export SOLACE_BROKER_USERNAME=admin
export SOLACE_BROKER_PASSWORD=admin
export SOLACE_BROKER_VPN=default
export SOLACE_IS_QUEUE_TEMPORARY=true

export OPENAI_MODEL_NAME=<LLM MODEL NAME>
export OPENAI_API_KEY=<LLM KEY>
export OPENAI_API_ENDPOINT=<LLM ENDPOINT>

export QDRANT_URL=<QDRANT CLUSTER URL>
export QDRANT_API_KEY=<QDRANT API KEY>
export QDRANT_COLLECTION=<A NAME FOR QDRANT COLLECTION>
export QDRANT_EMBEDDING_DIMENSION=1024
export DOCUMENTS_PATH=<PATH OF SOURCE DOCUMENTS IN LOCAL DISK>

Key Configuration Sections

• Scanner Configuration: Document source and monitoring settings
• Preprocessor Configuration: Text extraction and cleaning settings
• Splitter Configuration: Document chunking settings
• Embedding Configuration: Vector embedding settings
• Vector Database Configuration: Storage and retrieval settings
• LLM Configuration: Language model settings for augmentation
• Retrieval Configuration: Search parameters

Usage

Running the RAG System

solace-agent-mesh run

Querying the RAG System

Open the SAM UI on the browser.

Ingesting documents

(Option1): Store documents in a specific directory and configure the directory path in the rag.yaml file. After running SAM, the plugin ingests documents in background automatically.

(Option2): Open the SAM UI on the browser (by default http://localhost:8000), attach files to a query such as "ingest the attached document to RAG". This query persistently stores the attachments in file system and index them in vector database.

Retrieving documents

Use SAM UI on the browser (by default http://localhost:8000) or any other interfaces and send a query such as "search documents about and return a summary and referenced documents". It retrieves top similar documents and returns a summary of documents align with their original documents.