n8n Agent

n8n Agent AI is an intelligent, autonomous agent designed to build, verify, and update n8n workflows through natural language.

Instead of manually dragging and dropping nodes, users can simply describe the automation they need (e.g., "Create a workflow that fetches unread emails from Gmail and posts them to a Slack channel"). The agent intelligently queries a vector database of n8n node documentation, generates the workflow JSON, verifies its syntax, self-corrects using an LLM Critic, and streams the entire process back to the user in real-time.

*The n8n Agent AI streaming its internal thought process and generating workflows.*

*The fully deployed and syntax-verified workflow running inside n8n.*

Key Features

🧠 Intelligent Workflow Generation

Powered by the Vercel AI SDK and Gemini models, the agent doesn't just guess n8n JSON structures. It uses a Retrieval-Augmented Generation (RAG) approach to fetch the exact schema and parameter requirements for requested nodes, ensuring high-fidelity outputs.

🔄 Multi-Stage Self-Correction Pipeline

The agent operates using a robust pipeline to guarantee workflow validity:

1. Intent Parsing: Determines if the user is asking to create a new workflow or update an existing one.
2. Vector Retrieval: Queries PostgreSQL (via pgvector) to pull exact documentation for the requested integration nodes.
3. Generation: Drafts the initial n8n workflow JSON.
4. Syntax Verification: A deterministic engine checks for broken node connections, missing required fields, or cyclic loops.
5. LLM Critic: An independent LLM evaluates the generated workflow against the original user prompt. If the critic finds issues, it feeds the errors back into the generator for a self-correction loop.

⚡ Real-Time Thought Streaming

Gone are the days of staring at a loading spinner. The backend pipeline is fully streamed to the frontend via Server-Sent Events (SSE). Users can watch the agent's internal "thinking" process in real-time—seeing when it retrieves data, when it verifies syntax, and when it deploys.

📊 Full Observability

Deeply integrated with Laminar AI, every step of the agent's pipeline—from intent classification to final generation—is traced and logged. This ensures complete observability into latency, token usage, and LLM decision-making.

Tech Stack

• Frontend Framework: Next.js (App Router) & React
• AI Orchestration: Vercel AI SDK
• LLM Provider: Google Gemini (gemini-3.5-flash / gemini-2.5-flash)
• Database & Vector Search: Neon (PostgreSQL) with pgvector
• ORM: Drizzle ORM
• Observability: Laminar AI