SysReptor RAG Server

This directory contains a lightweight, fully self-contained Retrieval-Augmented Generation (RAG) server designed to provide context from historical, finished SysReptor Penetration tests. It exposes an easy-to-use REST API so that other AI agents can query the dataset locally or across a network.

Features

• Retrieval Only: It returns raw JSON chunks containing Executive Summaries, Technical Summaries, and Findings based on semantic vector similarity.
• Fast & Local: Powered by ChromaDB and the all-MiniLM-L6-v2 embedding model. It does not require any OpenAI/external API keys to run the database.
• Scrubbed Data: All data inside the vector database has had customer names, IP addresses, domains, and heuristic passwords scrubbed.

Deployment Instructions

To deploy this RAG server on another machine, follow these steps:

1. Move files: Copy this entire directory (specifically rag_server.py, ingest_data.py, scrubbed_rag_data.json, and requirements-rag.txt) to the target server.
2. Setup agents to use this RAG Move the agent_workflows to the local machine running the agent, so the agent can have the proper system prompt to know how to use this rag server
3. Setup virtual environment:

   python3 -m venv venv
   source venv/bin/activate

4. Install Dependencies:

   pip install -r requirements-rag.txt

5. Ingest Data: Create the local vector database instance in the ./chroma_db folder:

   python3 ingest_data.py

6. Run the Server: Start the FastAPI server. It is configured to listen on 0.0.0.0, meaning it will be accessible from any network interface on port 8000:

   python3 rag_server.py

   (Alternatively, use nohup python3 rag_server.py & to run in the background)

---

Agent Integration Guide (For AI Agents)

If you are an AI assistant trying to use the knowledge inside this RAG server, you can interface with it using standard HTTP requests (e.g., curl or requests library in Python).

Endpoint

POST http://<server_ip>:8000/query

Parameters

Field	Type	Required	Description
query	string	Yes	The natural language question you are trying to answer.
num_results	int	No	Number of chunks to return (Default: 5).
filter_type	string	No	Specifically scope results. Valid options: "finding", "executive_summary", "technical_summary". Defaults to returning everything combined.

Example Request (curl)

curl -X POST http://127.0.0.0:8000/query \
     -H "Content-Type: application/json" \
     -d '{
           "query": "How do we write an executive summary regarding ransomware?",
           "num_results": 3,
           "filter_type": "executive_summary"
         }'

Example Response

{
  "results": [
    {
      "content": "The objective of the assessment was to identify vulnerabilities within...",
      "metadata": {
        "project_id": "70ba26e4-2b69-4142-bfc5-77be90335990",
        "type": "executive_summary"
      },
      "distance": 0.42
    }
  ]
}

Tips for Agents

• If your initial query doesn't yield exactly what you want, try rephrasing the query string with different keywords. This uses Semantic Vector Search, so phrasing matters.
• Always check the type inside the metadata to ensure you are referencing the correct part of a report (e.g. don't paste an executive summary snippet into a finding template!).