installation.md

Installation

Last updated: 2025-11-26

Requirements

• Node.js ≥ 18, Docker + Docker Compose, Git, ~30 GB disk (large spaCy models + cache).
• Access to Llama Prompt Guard 2 model (Hugging Face, license acceptance).

Steps

1. Clone the repo and enter directory.
2. Download model:

./scripts/download-llama-model.sh   # saves to ../vigil-llm-models/

3. Set execute bits:

chmod +x install.sh scripts/download-llama-model.sh

4. Run installer:

./install.sh

• Creates .env, generates passwords, initializes ClickHouse, sets volumes, builds images.

5. Start all services:

docker-compose up -d

Verification

• Heuristics: curl http://localhost:5005/analyze -d '{"text":"hi","request_id":"t1"}'
• Webhook: curl http://localhost:5678/webhook/vigil-guard-2 -d '{"chatInput":"hi","sessionId":"demo"}'
• ClickHouse: curl http://localhost:8123/ -u user:pass
• Web UI: http://localhost/ui/

Important .env entries

• JWT_SECRET, CLICKHOUSE_PASSWORD, GF_SECURITY_ADMIN_PASSWORD
• GROQ_API_KEY (if required by LLM Safety Engine)
• Volume paths: vigil_data, clickhouse-data, grafana-data

Models

• Llama Prompt Guard 2 86M – stored outside repo: ../vigil-llm-models/Llama-Prompt-Guard-2-86M/ (gitignored).