🧠 Local AI Packaged

⚠️ Note: This project is a maintained and extended fork of coleam00/local-ai-packaged, which no longer appears to be actively maintained. This version includes up‑to‑date dependencies, bug fixes, and full automation scripts for setup and management.

---

🌍 Overview

Local AI Packaged provides a full self‑hosted AI environment using Docker Compose. It bundles everything you need for local LLM workflows, including:

• 🧠 Ollama — Local LLM inference (CPU / NVIDIA / AMD)
• 💬 Open WebUI — Chat interface for your models and n8n agents
• ⚙️ n8n — Workflow automation with Redis queue mode support
• 🌊 Flowise — Low-code AI chain builder
• 🧱 Supabase — Database, vector store, and auth system (optional)
• 📊 Langfuse — LLM tracing and observability
• 📦 Qdrant — Vector database for RAG
• 🕸️ Neo4j — Graph database
• 🔍 SearXNG — Web search engine for RAG pipelines
• 🔧 Unsloth — LLM fine-tuning studio (NVIDIA GPU required, ~20 GB image)
• 🔒 Caddy or SWAG — Reverse proxy with auto HTTPS (optional)

This version is designed for technical self‑hosters running the stack on a home server, NAS (e.g. Synology), or any Linux/Windows host.

---

⚙️ Prerequisites

• 🐳 Docker and Docker Compose v2+
• 🐍 Python 3.8+
• 💾 Git
• 💡 At least 16 GB RAM recommended

GPU (optional)

• NVIDIA: Install NVIDIA Container Toolkit
• AMD: ROCm runtime configured
• CPU only: Works fine, just slower

---

🚀 Quick Start

① Clone the repository

git clone https://github.com/gamersalpha/local-ai-packaged.git
cd local-ai-packaged

② Generate the .env automatically

python3 generate_env.py --yes --regen-sensitive

Pour choisir votre domaine de base (les sous-domaines de chaque service seront dérivés automatiquement) :

# Via l'option --domain
python3 generate_env.py --yes --regen-sensitive --domain home.example.com

# Ou en mode interactif (sans --yes), le script demandera le domaine
python3 generate_env.py --regen-sensitive

Features:

• Generates secure random secrets
• Detects GPU type (NVIDIA / AMD / CPU)
• Sets correct local paths for volumes
• Configures BASE_DOMAIN interactively or via --domain flag
• Expands BASE_DOMAIN into per-service hostnames automatically

③ Deploy the stack

Interactive setup wizard (recommended for first time):

python3 start_services.py --setup

Or direct launch:

python3 start_services.py --profile cpu --no-supabase --no-caddy

---

🧩 Managing the Stack

▶️ start_services.py

The main orchestrator with service selection, proxy detection, and interactive wizard.

python3 start_services.py [options]

Options

Option	Description
--setup	Interactive setup wizard with service picker
--profile [cpu|gpu-nvidia|gpu-amd]	Select hardware profile
--environment [private|public]	Network mode (localhost-only or LAN)
--services [names...]	Select specific services to deploy
--proxy [caddy|swag|none|auto]	Reverse proxy type (auto-detects SWAG)
--no-supabase	Skip Supabase
--no-caddy	Skip Caddy reverse proxy
--update	Pull latest Docker images before start
--dry-run	Preview configuration only

Selectable Services

Use --services to deploy only what you need. Dependencies are resolved automatically (e.g. n8n will also start postgres and redis).

# Deploy only n8n, Open WebUI, and Ollama
python3 start_services.py --services n8n openwebui ollama

# Deploy everything (default)
python3 start_services.py --services all

Available: n8n, openwebui, flowise, qdrant, neo4j, langfuse, searxng, ollama, unsloth

Examples

# Interactive wizard
python3 start_services.py --setup

# CPU, all services, no reverse proxy
python3 start_services.py --profile cpu --no-supabase --no-caddy

# NVIDIA GPU, public network with SWAG proxy
python3 start_services.py --profile gpu-nvidia --environment public --proxy swag

# Dry-run to preview what would be deployed
python3 start_services.py --services n8n openwebui ollama --dry-run

# Pull latest images and restart
python3 start_services.py --update

💡 What it does:

• Validates or creates your .env
• Clones Supabase's official Docker stack if missing
• Toggles Caddy and Supabase dynamically in docker-compose.yml
• Generates a new secret key for SearXNG
• Auto-detects SWAG reverse proxy and installs nginx configs
• Resolves service dependencies automatically
• Stops existing containers before redeploy
• Starts Supabase first (if enabled), then the Local AI stack

---

♻️ Update all services

./update_services.sh [profile]

Examples:

./update_services.sh cpu
./update_services.sh gpu-nvidia

This will stop all containers, pull latest images, and restart the stack.

---

🌐 Access Your Services

Service	Description	Default URL (private)
n8n	Workflow automation	http://localhost:5678
Open WebUI	Chat interface for LLMs	http://localhost:8080
Ollama	Local LLM API	http://localhost:11434
Flowise	Low-code AI builder	http://localhost:3001
Langfuse	LLM tracing dashboard	http://localhost:3000
SearXNG	Web search for RAG	http://localhost:8081
Neo4j	Graph database browser	http://localhost:7474
Qdrant	Vector database API	http://localhost:6333
PostgreSQL	Shared database	localhost:5433
Unsloth	LLM fine-tuning studio (~20 GB image, long pull)	http://localhost:8888
Supabase	DB, Auth & API (optional)	http://localhost:54323

In private mode, all ports are bound to 127.0.0.1. In public mode, they are accessible on all interfaces.

---

🌐 Domain Configuration

Set BASE_DOMAIN in your .env to auto-derive all service hostnames:

BASE_DOMAIN=home.example.com

This generates the following subdomains (à créer en tant que CNAME ou A record dans votre DNS) :

Sous-domaine	Service	Port interne
hub.BASE_DOMAIN	Landing page (dashboard)	8090
n8n.BASE_DOMAIN	n8n	5678
openwebui.BASE_DOMAIN	Open WebUI	8080
flowise.BASE_DOMAIN	Flowise	3001
langfuse.BASE_DOMAIN	Langfuse	3000
searxng.BASE_DOMAIN	SearXNG	8081
ollama.BASE_DOMAIN	Ollama	11434
qdrant.BASE_DOMAIN	Qdrant	6333
neo4j.BASE_DOMAIN	Neo4j	7474
unsloth.BASE_DOMAIN	Unsloth	8888
supabase.BASE_DOMAIN	Supabase	8000

Exemple : avec BASE_DOMAIN=home.example.com, créez un wildcard DNS *.home.example.com pointant vers l'IP de votre serveur, ou ajoutez chaque sous-domaine individuellement.

Override individual services:

N8N_HOSTNAME=custom-n8n.mydomain.com

---

🔒 Reverse Proxy

Caddy (built-in)

Enabled by default in public mode. Auto-generates Let's Encrypt certificates.

SWAG (auto-detected)

If you already run SWAG on your server (Synology, Unraid, etc.), start_services.py detects it automatically and installs nginx proxy configs from the swag/ directory.

python3 start_services.py --proxy swag
# or let it auto-detect:
python3 start_services.py --proxy auto

---

⚡ n8n Queue Mode

For heavy workflows, enable Redis-backed queue mode with separate worker containers:

# In .env:
N8N_EXECUTIONS_MODE=queue

Then start with the worker profile:

python3 start_services.py --profile cpu
docker compose -p localai --profile n8n-worker up -d

---

🗄️ Database Isolation

Each service uses its own PostgreSQL database to prevent schema collisions:

Service	Database
n8n	n8n
Langfuse	langfuse
Flowise	flowise
Supabase	postgres

Databases are auto-created on first start via postgres/init/01-create-databases.sql.

---

🛠️ Troubleshooting

Supabase fails to start

• Check that the folder supabase/ was created automatically
• Delete it and rerun: python3 start_services.py
• Ensure .env contains POOLER_DB_POOL_SIZE=5

GPU not detected

• Ensure the NVIDIA Container Toolkit or ROCm is installed correctly
• Fallback to CPU: python3 start_services.py --profile cpu

Ports already in use

• Check what's using the port: netstat -tlnp | grep <port>
• Edit docker-compose.override.private.yml to change exposed ports

Unsloth image takes forever to pull

• The unsloth/unsloth:latest image is ~20 GB (CUDA + PyTorch). First pull can take 10-30 minutes depending on your connection.
• You can pull it in the background: docker pull unsloth/unsloth:latest
• Deploy all other services first without Unsloth, then add it later with: docker compose -p localai --profile gpu-nvidia -f docker-compose.yml -f docker-compose.override.private.yml up -d unsloth

Container healthcheck failing

• All healthchecks use 127.0.0.1 (not localhost) to avoid IPv6 issues
• Check logs: docker logs <container_name>

---

🧾 Project Structure

.
├── docker-compose.yml                 # Main orchestration file
├── docker-compose.override.private.yml # Localhost-only port bindings
├── docker-compose.override.public.yml  # LAN-accessible port bindings
├── start_services.py                  # Smart deployment launcher
├── generate_env.py                    # .env generator with GPU detection
├── update_services.sh                 # Container update helper
├── Caddyfile                          # Caddy reverse proxy config
├── .env.example                       # Environment template
├── n8n_pipe.py                        # Open WebUI → n8n integration pipe
├── postgres/
│   └── init/                          # Database init scripts
├── swag/                              # SWAG proxy-conf templates
├── n8n/
│   └── backup/                        # Pre-built n8n workflows
├── flowise/                           # Flowise chatflows & custom tools
├── searxng/                           # SearXNG configuration
├── supabase/                          # Auto-cloned (gitignored)
├── shared/                            # Shared data volume (gitignored)
└── neo4j/                             # Neo4j data (gitignored)

---

📋 Roadmap / TODO

🔜 Planned Features

• Landing page dashboard — Page d'accueil Cyberpunk 2077 avec icônes, descriptions et liens dynamiques vers chaque service installé (hub.BASE_DOMAIN / port 8090)
• Environment-based versioning — Mode prod (versions pinnées) vs recette (latest) avec granularité par service. Permettre --env prod / --env recette dans start_services.py
• Supabase rework — Refaire l'intégration Supabase pour plus de fiabilité et de modularité (clone, .env partagé, sélection de services)
• Monitoring & alerting — Intégration Prometheus/Grafana pour surveiller l'état des services
• Backup automation — Script de backup automatique des volumes Docker et bases de données
• Multi-node support — Support Docker Swarm ou Kubernetes pour déploiement multi-serveur

✅ Recently Completed

• Selective service deployment (--services)
• Interactive setup wizard (--setup)
• SWAG reverse proxy auto-detection and config generation
• Global BASE_DOMAIN with per-service hostname derivation
• n8n v2 + Redis queue mode + worker profile
• Database isolation (separate PostgreSQL databases per service)
• Healthchecks on all services (IPv4-safe)
• Unified logging with rotation on all containers
• Dynamic update_services.sh with profile argument
• Redis authentication (--requirepass)

---

📜 License

Licensed under the Apache 2.0 License. See LICENSE for details.

---

Built and maintained with ❤️ for the self‑hosting community.