🚀 Enterprise AI-powered n8n workflow builder with CRAG-enhanced generation.
🔥 CRAG Pipeline: Jina Reranker + Mistral-Nemo Judge for superior retrieval.
| Feature | Description |
|---|---|
| 🧠 AI Workflow Builder | Generate n8n workflows from natural language (Gemini 3 Pro Preview) |
| 🔍 RAG-Powered | 36,166 workflow vectors for semantic retrieval |
| ⚡ Diff-Based Updates | Safe add/update/remove/reconnect actions (no canvas overwrites) |
| 🎨 Embedded n8n | Full n8n canvas with custom AI Builder panel |
| 🤖 Custom Model | Qwen 3 Coder 14B fine-tuned on 21,925 n8n workflow examples |
┌─────────────────────────────────────────────────────────────────┐
│ Kaelux Automate │
├─────────────────┬─────────────────┬─────────────────────────────┤
│ Next.js 15 │ n8n Canvas │ AI Pipeline │
│ (Port 3000) │ (Port 5678) │ │
├─────────────────┴─────────────────┤ │
│ │ ┌───────────────────────┐ │
│ ┌─────────────────────────┐ │ │ Gemini 3 Pro Preview │ │
│ │ PostgreSQL 16 │ │ │ (Primary LLM) │ │
│ │ (Port 5433) │ │ └───────────┬───────────┘ │
│ └─────────────────────────┘ │ │ │
│ │ ┌───────────▼───────────┐ │
│ ┌─────────────────────────┐ │ │ Qwen 3 Coder 14B │ │
│ │ Qdrant Cloud (36k RAG) │ │ │ (Custom Fine-Tuned) │ │
│ └─────────────────────────┘ │ └───────────────────────┘ │
└───────────────────────────────────┴─────────────────────────────┘
| Component | Version | Purpose |
|---|---|---|
| Next.js | 15.x | API routes, Prisma/Postgres, AI orchestration |
| React | 19.x | Frontend framework |
| n8n | 1.x (forked) | Vue-based workflow canvas with AI Builder panel |
| PostgreSQL | 16.x | Primary data store |
| Prisma | 5.x | Database ORM |
| Qdrant | Cloud | Vector index for 36,166 workflow exemplars |
| Gemini | 3 Pro Preview | Primary LLM for workflow generation |
| Qwen 3 Coder | 14B (training) | Custom fine-tuned model for n8n workflows |
- Docker + Docker Compose v2
- Node.js 22.x / pnpm 9.x
- Gemini API key (3 Pro Preview)
# 1. Clone and configure
git clone https://github.com/Ker102/Kaelux-Automate.git
cd Kaelux-Automate
cp .env.example .env
cp .env.local.example .env.local
# 2. Add your API keys to .env.local
# GEMINI_API_KEY=your-key-here
# 3. Build n8n with AI Builder
cd n8n
pnpm install
pnpm build:n8n && pnpm build:docker
cd ..
# 4. Start services
docker compose -f docker-compose.dev.yml up -d
# 5. (Optional) Seed RAG database
docker compose exec app npm run seed:qdrant| Service | URL |
|---|---|
| Next.js App | http://localhost:3000 |
| n8n Canvas | http://localhost:5678 |
| Qdrant REST | http://localhost:6333 |
| PostgreSQL | localhost:5433 |
The AI builder uses Gemini 3 Pro Preview for generation and structured diff actions:
// Example API response
{
"workflow": { /* n8n workflow JSON */ },
"actions": [
{ "type": "add", "node": {...} },
{ "type": "update", "nodeId": "123", "changes": {...} },
{ "type": "connect", "from": "A", "to": "B" }
]
}Endpoints:
POST /api/ai/workflow- Generate workflow from promptGET /api/ai/prompts- List prompt templates
We're fine-tuning Qwen 3 Coder 14B specifically for n8n workflow generation:
| Metric | Value |
|---|---|
| Base Model | Qwen 3 Coder 14B |
| Training Framework | Unsloth (QLoRA) |
| Training Examples | 21,925 (cleaned) |
| Vector Dimensions | 768 (M2-BERT) |
| Training Platform | GCP / Colab |
cd n8n && pnpm build:n8n && pnpm build:docker && cd ..
docker compose -f docker-compose.dev.yml up -d n8ndocker compose exec app npx prisma migrate deploy
docker compose exec app npx prisma generate- Dependabot monitors npm and GitHub Actions
- CodeQL scans on pushes and PRs
- SECURITY.md documents disclosure process
See CONTRIBUTING.md for guidelines. All changes require PRs with tests/docs.
Copyright © Kaelux. See repository license for details.
- n8n Automation Atlas - Source of 131k+ workflows powering the RAG
- HuggingFace Dataset - Training data for fine-tuning