Chatbot

This project provides:

• BookStack page export to PDF
• Vector store indexing (Chroma)
• A RAG chatbot API (OpenAI-compatible endpoints)
• An Open WebUI interface connected to the RAG API

Prerequisites

• Python 3.11+
• Docker + Docker Compose
• OpenAI API key with active billing/quota
• BookStack API token (for export only)

Environment setup

Create .env from .env.example and fill values:

OPENAI_API_KEY=sk-proj-your-api-key
MODEL_LLM=yout_model
MODEL_EMBEDDING=your_embedding_model

BOOKSTACK_URL=your_url
BOOKSTACK_TOKEN_ID=your_bookstack_token_id
BOOKSTACK_TOKEN_SECRET=your_bookstack_token_secret

CHATBOT_API_BIND_IP=IP
CHATBOT_API_HOST_PORT=PORT
CHATBOT_API_INTERNAL_PORT=PORT
ANYTHINGLLM_BIND_IP=IP
ANYTHINGLLM_HOST_PORT=PORT

Python setup

python -m venv .venv
source .venv/bin/activate
pip install -r requirements.txt

1) Export BookStack pages as PDFs

python export_pages.py

PDF files are written to exports/.

2) Build / refresh vector store

python reload_vector_store.py

This indexes documents from exports/ into vector-store/.

3) Run chatbot API + Open WebUI

docker compose up --build

Access:

• AnythingLLM: http://localhost:${ANYTHINGLLM_HOST_PORT} (default: 3001)
• Chatbot API health: http://localhost:${CHATBOT_API_HOST_PORT}/health (default: 8000)

Open WebUI is configured to call the local chatbot backend through OpenAI-compatible routes:

• GET /v1/models
• POST /v1/chat/completions

4) Expose AnythingLLM on internet (secure minimal setup)

Files:

• docker-compose.anythingllm.secure.yml
• deploy/Caddyfile

This stack puts Caddy in front of AnythingLLM:

• HTTPS with automatic TLS certificates
• HTTP Basic Auth at proxy level
• AnythingLLM not exposed directly (internal only)

Steps:

1. Create env file:

docker compose up

Security notes:

• Keep AnythingLLM built-in auth enabled (multi-user recommended for internet exposure).
• Disable public signup inside AnythingLLM unless explicitly needed.
• Keep OPENAI_API_KEY only inside server env, never in frontend code.

5) Automatize the vector store reload :

Create a job to automatically update the vectore store weekly : CRON_TZ=Europe/Paris 0 7 * * 1 chatbot/reload_job.sh >> chatbot/reindex.log 2>&1

Optional: notebooks

pip install jupyter ipykernel
python -m ipykernel install --user --name chatbot-venv --display-name "Python (chatbot-venv)"
jupyter notebook