Demos & Synthetic Data Factory User Guide

This guide provides engineering runbooks for demonstrating Caber zero trust policy enforcement, subdocument characterization, multi agent delegation, and deterministic synthetic data generation using CaberDemos and CaberDataFactory.

Figure 1: Deterministic Synthetic Data Generation Feeding CaberDemos and Benchmark Harnesses

## Part 1: `CaberDemos` Installation & Operation User Guide

1.1 Overview & Catalog of the 4 Active Demo Scenarios

Located at CaberDemos, CaberDemos consolidates product demonstrations into a unified CLI (caber-demo), Docker Compose overlay stack, and live scenario runner.

Demo ID	Scenario Name	Primary Ports	Key Capabilities Demonstrated
agentic-tools	Genuine Agentic Retrieval with LangGraph & Milvus	8082 (MCP), 8083 (Chat UI), 19530 (Milvus)	A LangGraph autonomous agent dynamically selects among four MCP tools and issues vector queries, with synchronous Caber policy enforcement at the LangChain BaseTool boundary.
hipaa-ner	Healthcare NER & Surgical PHI Redaction	8050 (Console), 8051 (Streamlit)	Evaluates clinical consultation records using ONNX GLiNER models, demonstrating synchronous redaction of protected health information before ingestion into vector indices.
agentgov-mimic	Multi Agent Delegation & Principal Identity Attribution	8050 (Audit), 8054 (Gateway)	Orchestrator Agent delegates to subagents across trust boundaries on behalf of specific human Principals, validating that identity tokens propagate correctly to prevent confused deputy leaks.
graphrag	Enterprise GraphRAG & Access Control Leakage Defense	7474 (Neo4j), 8050 (Console)	Evaluates knowledge graph traversals across interconnected nodes, proving that access control policies prevent sensitive node traversal leaks.

1.2 Prerequisites & Workstation Virtual Environment Setup

Before launching any demo scenario, configure your workstation environment:

1. Prerequisite Services:

   • Docker Engine and Docker Compose v2 (required for local OpenSearch, S3 compatible store such as MinIO or VersityGW, Neo4j, PostgreSQL, and Milvus services).
   • Python 3.10, 3.11, 3.12, or 3.13.

2. Step 1: Create Local Python Virtual Environment:

cd CaberDemos
python3 -m venv .venv
source .venv/bin/activate

3. Step 2: Install Package Dependencies:

.venv/bin/pip install --upgrade pip
.venv/bin/pip install -e ".[all]"
.venv/bin/pip install -e "../PyCaber[auto]"

4. Step 3: Configure Environment Variables: Copy .env.example to .env and set your OpenAI API key and local service credentials:

cp .env.example .env
# Edit .env to supply OPENAI_API_KEY

1.3 Running Demos via the caber-demo CLI

The caber-demo runner (demo.py) orchestrates docker compose stacks, dataset loading, and service health checks:

# List available demo scenarios
.venv/bin/python demo.py list

# Launch the agentic tools demo
.venv/bin/python demo.py start agentic-tools

# Check running container health and ports
.venv/bin/python demo.py status

# Stop demo and clean up containers
.venv/bin/python demo.py stop agentic-tools

Part 2: CaberDataFactory Synthetic Generation Runbook

Located at CaberDataFactory, this tool generates deterministic corporate corpora, organizational charts, access control lists, and version cluster topologies for testing.

Figure 2: Deterministic Scenario Driven Corpus Generation with Version Cluster Topologies

### 2.1 Virtual Environment Setup for `CaberDataFactory` ```bash cd CaberDataFactory python3 -m venv .venv .venv/bin/pip install -e ".[all]" ``` ### 2.2 CLI Usage & Scenario Generation ```bash # Generate deterministic corporate M&A scenario with seed 42 .venv/bin/python -m datafactory generate --scenario m_and_a --seed 42 --out ./output_corpus

Validate generated document metadata against CONFIG_SCHEMA.json

.venv/bin/python -m datafactory validate --dir ./output_corpus