Architecture#

QAnswer can be self-hosted on your own infrastructure via Docker Compose, giving you full control over data residency, networking, and scaling. This page covers the system architecture, requirements, and a quick start for deploying the bundle.

Architecture of the application#

Here is a diagram explaining the overall architecture of the QAnswer application:

The global architecture of QAnswer is composed of 3 parts:

• Nginx WAF — responsible for external routing, Web Application Firewall, and TLS/SSL management.
• QAnswer Core — the functional part of QAnswer providing the required functionalities.
• QAnswer Monitoring — responsible for monitoring the status of QAnswer.

QAnswer Core is composed of the following services:

• Traefik — load balancer for internal traffic routing;
• QAnswer Frontend — the web GUI of the application;
• QAnswer Doc — the documentation site you're reading right now;
• QAnswer Python Backend — entrypoint of the backend, responsible for security, authentication, authorization, AI models, and indexing;
• QAnswer LLM — the large language model serving layer;
• QAnswer Embedder — the embedding model used for semantic search;
• QAnswer Transcription — speech-to-text;
• Clamav — checks uploaded files for viruses;
• Selenium — web scraping;
• PostgreSQL — SQL data storage;
• Valkey — key/value store for caching and queuing;
• ElasticSearch — document/vector storage for retrieval.

Requirements#

Hardware

• 16 GB RAM
• 200 GB disk
• 48 GB vRAM GPU

Software

• Any Linux distribution
• Docker version 20 or newer
• Docker Compose
• The nvidia-container-toolkit installed, to use NVIDIA GPUs in a container
• A licence key, plus a username and access token (provided by us)

(This supports 7 to 13 concurrent users depending on RAG or Chat workload. Contact us at info@the-qa-company.com for help sizing your infrastructure.)

Quick start#

Once you have your licence key and registry credentials, deploying the bundle takes a few steps:

1. Clone the bundle repository

git clone -b main https://gitlab.the-qa-company.com/qanswer-app/qanswer-bundle.git
cd qanswer-bundle

2. Set your credentials

export QANSWER_KEY=YOUR_KEY
export HUGGINGFACE_TOKEN=YOUR_HF_KEY

YOUR_KEY is your licence key; YOUR_HF_KEY is the Hugging Face token we provide.

3. Log in to the image registry

docker login -u $QANSWER_REGISTRY_USER -p $QANSWER_REGISTRY_PASSWORD \
  https://registry.gitlab.the-qa-company.com:443/v2/qanswer-app

QANSWER_REGISTRY_USER and QANSWER_REGISTRY_PASSWORD are the docker registry credentials we provide.

4. Download all images

./init.sh

5. Start QAnswer

./deploy-scripts/up.sh

# or manually:
docker compose up -d

6. Open your browser at http://localhost:4567/

7. The default login is username: admin / password: 123456