Sign inSign up

thegmpteam/n8n-ai-stack

By thegmpteam

•Updated 8 days ago

Complete n8n AI automation stack with Ollama LLM and ComfyUI image/video generation.

Image
0

761

thegmpteam/n8n-ai-stack repository overview

⁠n8n AI stack — Docker-only CLI

Terminal control for exactly six services: n8n-table-ui, n8n, Ollama, PocketTTS, SearXNG and ComfyUI. The benign VirusGPT vgctl.py control/diagnostic foundation is ported to JavaScript.

Only prerequisite: the Docker CLI and a working Linux-container Docker daemon. Node, npm, Python, Git, GitHub CLI, curl and the host Compose plugin are not required — the image contains Node, Git, Docker CLI, Buildx and Compose.

This is an experimental beta, not a completed video-production system.

⁠Quick start

docker pull thegmpteam/n8n-ai-stack:beta

Linux or macOS Terminal, from an empty directory you own:

docker run --rm -it \
  --add-host host.docker.internal:host-gateway \
  --mount type=bind,source=/var/run/docker.sock,target=/var/run/docker.sock \
  --mount "type=bind,source=$PWD,target=/workspace" \
  thegmpteam/n8n-ai-stack:beta install

Windows PowerShell (Docker Desktop must run Linux containers):

docker run --rm -it `
  --add-host host.docker.internal:host-gateway `
  --mount 'type=bind,source=/var/run/docker.sock,target=/var/run/docker.sock' `
  --mount "type=bind,source=$((Get-Location).Path),target=/workspace" `
  thegmpteam/n8n-ai-stack:beta install

$PWD must be the absolute installation parent directory. Installation defaults to /workspace/n8n-ai-install, persisted as ./n8n-ai-install on your host. On rootless Docker use your actual daemon socket source, still mounted at /var/run/docker.sock. Remote Docker contexts are unsupported.

⁠Commands

Repeat the docker run command, replacing the final install with:

doctor
doctor --fix
run pockettts
stop pockettts
status
status comfyui
logs n8n
models list
workflows import --url ORIGIN --expected-url ORIGIN --credentials FILE --journal FILE --apply

Optional POSIX and PowerShell launchers are bundled at launch/n8n-stack.sh and launch/n8n-stack.ps1. Set N8N_STACK_IMAGE to an immutable digest to pin the image.

Default installation directory: ./n8n-ai-install (or N8N_STACK_DIR). --fix never installs system software, kills ports or restarts services.

⁠Installer behaviour

install asks which of the six services to include, Compose project, listener IP, advertised n8n/browser hostname, timezone, each selected host port, Comfy browser endpoint, external replacements, GPU UUIDs, Pocket CPU/thread budgets and the enhancement model tag.

  1. Type yes to prepare private state and copy hash-verified UI/Pocket source snapshots. No Git fetch or runtime private-repository authentication occurs. Cancelling before consent writes no files.
  2. Enter optional runtime secrets only in the masked interactive terminal. Review, encryption and SearXNG secrets are generated privately.
  3. Real bundled Compose validates the plan. A separate type yes authorizes Docker build/pull/start. Declining leaves preparation only.

No workflow import/activation or model-download job runs automatically. No existing production Compose project or data is adopted, stopped or deleted.

CPU Ollama and CPU Pocket require no host language runtime. Local Comfy is opt-in and CUDA-only; macOS and unsupported hosts must exclude it and supply an external Comfy endpoint.

⁠Persistent host paths

n8n-ai-install/
  stack.json               owned identity, answers, exact source lock
  compose.json             local build contexts and bind definitions
  compose.daemon.json      long-syntax binds using actual daemon host paths
  .env                     runtime keys/tokens (private)
  sources/ui/              hash-verified pinned runtime snapshot
  sources/pockettts/       hash-verified pinned runtime snapshot
  data/                    persistent per-service binds

⁠Tags

TagMeaning
betaCurrent published build (there is no latest tag yet)
sha-<commit>Immutable per-commit build

Platforms: linux/amd64, linux/arm64.

⁠Security boundary

Mounting the Docker socket gives this CLI control of the host/daemon, equivalent to administrator access. Run only trusted, reviewed images. Never expose the daemon over unauthenticated TCP and never loosen socket permissions to solve access errors.

The image uses explicit COPY allowlists and a restrictive .dockerignore: no .git, .env, keys, private audio, media, model caches or installed data.

⁠License

MIT (VirusGPT foundation attribution preserved) / ISC source license.

Tag summary

Content type

Image

Digest

sha256:087d50abd…

Size

117.3 MB

Last updated

8 days ago

docker pull thegmpteam/n8n-ai-stack