EZ-CorridorKey Docker Guide

Run the GUI in a browser via noVNC, with persistent volumes for projects, models, and runtime data.

Prerequisites

• Docker Engine + Docker Compose plugin
• NVIDIA users: working NVIDIA Container Toolkit (for GPU service)

Start

From the docker/ directory:

CPU:

docker compose up -d corridorkey-cpu --build

GPU:

docker compose --profile gpu up -d corridorkey-gpu --build

First run will take several minutes (installs Python dependencies + downloads models). Subsequent starts are fast.

To see logs during first run: omit -d to run in the foreground.

Access the app

All three endpoints are bound to 127.0.0.1 only (never exposed to your LAN) and require authentication. The password for every endpoint is the repository name: EZ-CorridorKey.

Endpoint	URL	User	Password
Web UI (noVNC)	http://localhost:6080	—	EZ-CorridorKey
Upload UI (file browser)	http://localhost:6081	admin	EZ-CorridorKey
Raw VNC (optional)	localhost:5900	—	EZ-CorridorKey

Changing the password: edit the VNC password in supervisord.conf (x11vnc -passwd ...) and the filebrowser admin creds in the same file (filebrowser users add admin <new-password>), then recreate the container with docker compose --profile gpu up -d corridorkey-gpu --force-recreate. The filebrowser user only gets created on the first boot of a fresh .filebrowser.db; if you change the password after first boot you'll also need to docker exec in and run the filebrowser users update admin --password <new> command, or wipe the volume.

Upload files

1. Open http://localhost:6081
2. Upload clips into ClipsForInference
3. In the app, import from that folder

Uploads are persisted in the corridorkey_uploads volume.

Stop / restart

Stop services (keep all volumes/data):

docker compose stop

Start again:

docker compose start

Stop and remove containers (keep volumes/data):

docker compose down

Update project

git pull
docker compose --profile gpu up -d corridorkey-gpu --build --force-recreate

For CPU, replace service with corridorkey-cpu.

Environment changes

If you change environment values in docker-compose.yml, recreate the service:

docker compose --profile gpu up -d corridorkey-gpu --force-recreate

The startup script re-checks install/model env flags on every start and applies changes when needed.

Logs

GPU service:

docker compose logs -f corridorkey-gpu

CPU service:

docker compose logs -f corridorkey-cpu

Volumes (persistent)

• corridorkey_install → app source + .venv
• corridorkey_projects → project data
• corridorkey_models_* → model checkpoints
• corridorkey_hf_cache → Hugging Face cache
• corridorkey_config → app-level config
• corridorkey_logs → logs
• corridorkey_uploads → uploaded files

Security note

All three service ports (6080, 6081, 5900) bind to 127.0.0.1 only, so they are never reachable from your LAN or the internet out of the box. Every endpoint also requires a password (EZ-CorridorKey by default — see the table above). If you need remote access, tunnel over SSH rather than re-binding the ports to 0.0.0.0:

ssh -L 6080:localhost:6080 -L 6081:localhost:6081 user@your-host

Before publishing a Docker image publicly or handing the default container to untrusted users, change the password in supervisord.conf. The default is deliberately well-known — it exists so local/single-user setups don't have to guess a randomized password, not as real access control against a determined attacker.