AI Chat Stack

Run MD Planner alongside Ollama (LLM), Chatterbox (TTS), SearXNG (search), and
Caddy (reverse proxy) as a single composed stack.

Prerequisites

Docker, Docker Compose v2. Optional: NVIDIA GPU + Container Toolkit for
hardware-accelerated Ollama and Chatterbox.

Start the stack

docker compose -f deploy/docker-compose.yml up -d

Open http://localhost:8080. All services are accessible through Caddy.

Service	External URL	Notes
mdplanner	http://localhost:8080	Main application
Ollama API	internal only (ollama:11434)	Configure in AI Chat settings
Chatterbox	http://localhost:8080/tts	Routed via Caddy
SearXNG	http://localhost:8080/search	Routed via Caddy

AI Chat configuration

In mdplanner, navigate to AI Chat and open the Config panel:

• Ollama URL: http://ollama:11434 (Docker internal hostname)
• Chatterbox URL: http://localhost:8080 (through Caddy at /tts)
• SearXNG URL: http://localhost:8080 (through Caddy at /search)

GPU support

GPU is disabled by default. To enable NVIDIA GPU for Ollama and/or Chatterbox,
uncomment the deploy block in deploy/docker-compose.yml for the relevant
service. Requires
NVIDIA Container Toolkit.

Volumes

Volume	Contents
mdplanner-data	Project files (/data in container)
ollama-data	Downloaded LLM models
huggingface-cache	Chatterbox model weights
searxng-data	SearXNG cache

Voice reference WAV files for Chatterbox are bind-mounted from
deploy/chatterbox/references/. Add .wav files there before starting the
stack.

Security

Change the secret_key in deploy/searxng/settings.yml before any
public-facing deployment. The default value is a placeholder.

Management

make -f deploy/Makefile stack-up
make -f deploy/Makefile stack-down
make -f deploy/Makefile stack-logs