AI-powered n8n workflow engineering studio.
Describe your automation in plain English → get production-ready n8n workflow JSON with validation, documentation, and one-click deploy.
- Multi-agent pipeline — 6 specialized agents (Enhancer → Analyst → Architect → Builder → Repair → Designer) generate workflows step-by-step
- 106 specialist knowledge packs — domain experts for SEO, CRM, databases, AI/LLM, messaging, e-commerce, and more activate dynamically per prompt
- AI Agent pattern — generates modern n8n LangChain Agent workflows with proper sub-node connections (not raw HTTP)
- 20 validation rules — structural, expression, configuration, and security checks catch errors before you import
- Self-healing JSON — 3-stage repair (mechanical fixes → LLM-assisted → retry) handles malformed LLM output
- Coverage audit — detects missing services and auto-extends incomplete workflows
- Interactive canvas — pan, zoom, drag nodes, click to inspect parameters
- Chat-driven repair — describe an n8n error in the sidebar chat → workflow updates in-place
- Session persistence — every generation saved, rename/delete sessions, pick up where you left off
- Per-agent model selection — configure which Claude model each agent uses (Haiku for speed, Opus for reasoning)
- n8n instance management — connect, test, and deploy directly to your n8n servers
- Sticky note documentation — every workflow ships with visual section labels for the n8n canvas
- 10 prompt templates — pre-built recipes for common automation patterns
- Auto-fallback — if a model 503s, automatically retries with a lighter model
User prompt
↓
[Prompt Enhancer] — compresses + clarifies intent
↓
[Requirements Analyst] — extracts triggers, systems, steps
↓
[Workflow Architect] — selects nodes, maps connections
↓
[Skill Index] — picks top 7 specialists from 106 available
↓
[n8n Builder] — generates workflow JSON with specialist knowledge
↓
[Coverage Audit] — checks all services present, extends if missing
↓
[Validator] — 20 rules (structural, expression, config, security)
↓
[Repair Agent] — fixes validation failures (up to 3 attempts)
↓
[Designer] — adds sticky notes and section documentation
↓
Ready to deploy
See AGENTS.md for the full agent registry (106 specialists across 20 categories).
# Clone
git clone https://github.com/Xelson431/flowfactory-ai.git
cd flowfactory-ai
# Install
npm install
# Configure
cp .env.example .env.local
# Edit .env.local — set LLM_API_KEY (required)# Run
npm run dev
# Open http://localhost:3000/studio| Variable | Required | Description |
|---|---|---|
LLM_API_KEY | ✅ | API key for the LLM gateway |
LLM_GATEWAY_URL | No | OpenAI-compatible endpoint (default: Kiro Gateway) |
FLOWFACTORY_API_KEY | No | If set, all API routes require this key in x-api-key header |
ENCRYPTION_KEY | No | AES-256-GCM key for encrypting n8n API keys at rest |
DATABASE_URL | No | PostgreSQL connection (default: file-based) |
NEXTAUTH_SECRET | No | Session encryption (auth disabled by default) |
src/
├── app/ # Next.js App Router pages
│ ├── studio/ # Main IDE workspace (canvas + chat)
│ ├── templates/ # Prompt template library
│ ├── history/ # Generation archive
│ ├── instances/ # n8n connection management
│ └── api/ # API routes (generate, repair, deploy, sessions)
├── server/ # Backend logic
│ ├── orchestrator.ts # Multi-agent pipeline
│ ├── llm.ts # LLM provider with auto-fallback
│ ├── validation.ts # 20-rule validation engine
│ ├── json-repair.ts # Self-healing JSON parser
│ ├── coverage-audit.ts # Service coverage checker
│ ├── designer.ts # Sticky note documentation agent
│ ├── prompt-enhancer.ts # Input compression + clarification
│ ├── node-registry.ts # Built-in vs community node detection
│ ├── models.ts # Per-agent model configuration
│ └── specialists/ # 106 domain specialist knowledge packs
│ ├── core.ts # Webhook, HTTP, DB, AI, messaging, etc.
│ ├── seo.ts # 10 SEO specialists
│ ├── research.ts # Market/niche/competitor research
│ ├── llm-endpoints.ts # 9 LLM provider API references
│ ├── node-catalog.ts # 100+ services → node type mapping
│ ├── meta.ts # Always-on: linker, validator, best practices
│ └── ... # Google, Microsoft, CRM, PM, e-commerce, etc.
├── schemas/ # Zod schemas for all data types
└── types/ # Branded domain types
npm run test# Run all tests (47 passing)
npm run build # Production buildFlowFactory works with any OpenAI-compatible API. Set LLM_GATEWAY_URL to:
| Provider | URL |
|---|---|
| OpenAI | https://api.openai.com/v1/chat/completions |
| Anthropic (via proxy) | Your proxy URL |
| Groq | https://api.groq.com/openai/v1/chat/completions |
| Together.AI | https://api.together.xyz/v1/chat/completions |
| OpenRouter | https://openrouter.ai/api/v1/chat/completions |
| Ollama (local) | http://localhost:11434/v1/chat/completions |
- No secrets hardcoded — all credentials via environment variables
- API keys never logged or returned to the browser
- Generated workflows use
{{CRED:name}}placeholders (never raw secrets) - SQL injection prevention enforced in generated database nodes
- Prompt injection detection in user inputs
- HMAC signature verification patterns for webhook security
- Rate limiting — 30 requests/minute per IP on all API routes
- Optional API key auth — set
FLOWFACTORY_API_KEYto lock down endpoints - AES-256-GCM encryption — n8n instance API keys encrypted at rest
- Body size limits — 2MB max request body
- Prototype pollution protection — whitelist-only field updates on mutable routes
POST /api/workflows/generate
Content-Type: application/json
{
"projectId": "uuid",
"prompt": "When a webhook receives an order...",
"style": "robust",
"models": { "architect": "claude-sonnet-4.6" }
}
# Returns: SSE stream of generation eventsPOST /api/generate
# Same body → returns { workflow, validation, id }POST /api/repair
{ "workflowJson": {...}, "errorMessage": "Node X failed with 401" }
# Returns: { workflow, validation }POST /api/deploy
{ "instanceId": "uuid", "workflowJson": {...} }
# Pushes as inactive draft to n8nSee CONTRIBUTING.md for development setup and guidelines.
MIT — use it however you want.
- n8n — the workflow automation platform
- n8n-mcp — live node documentation
- Anthropic Claude — powers the generation agents
- ui-ux-pro-max — design intelligence