n8n¶

Self-hosted workflow automation platform with AI agent capabilities.

Overview¶

400+ integrations (Slack, email, GitHub, databases, HTTP APIs)
AI nodes: AI Agent, Text Classifier, Information Extractor, Summarizer
Native Ollama integration for chat and embedding models
Visual workflow builder with branching, loops, error handling
Webhook triggers, scheduled runs, event-driven workflows
Fair-code license (free self-hosted, source-available)

Installation¶

Docker Compose (AI Starter Kit)¶

Full stack with n8n, Ollama, Qdrant, and PostgreSQL based on the official n8n-io/self-hosted-ai-starter-kit:

# docker-compose.yml
services:
  n8n:
    image: n8nio/n8n:latest
    restart: unless-stopped
    ports:
      - "5678:5678"
    environment:
      - N8N_ENCRYPTION_KEY=${N8N_ENCRYPTION_KEY}
      - DB_TYPE=postgresdb
      - DB_POSTGRESDB_HOST=postgres
      - DB_POSTGRESDB_PORT=5432
      - DB_POSTGRESDB_DATABASE=n8n
      - DB_POSTGRESDB_USER=n8n
      - DB_POSTGRESDB_PASSWORD=${POSTGRES_PASSWORD}
      - N8N_DIAGNOSTICS_ENABLED=false
      - N8N_PERSONALIZATION_ENABLED=false
    volumes:
      - n8n-data:/home/node/.n8n
    depends_on:
      postgres:
        condition: service_healthy

  postgres:
    image: postgres:16-alpine
    restart: unless-stopped
    environment:
      - POSTGRES_USER=n8n
      - POSTGRES_PASSWORD=${POSTGRES_PASSWORD}
      - POSTGRES_DB=n8n
    volumes:
      - postgres-data:/var/lib/postgresql/data
    healthcheck:
      test: ["CMD-SHELL", "pg_isready -U n8n"]
      interval: 5s
      timeout: 5s
      retries: 5

  qdrant:
    image: qdrant/qdrant:latest
    restart: unless-stopped
    ports:
      - "6333:6333"
    volumes:
      - qdrant-data:/qdrant/storage

volumes:
  n8n-data:
  postgres-data:
  qdrant-data:

# Generate encryption key
echo "N8N_ENCRYPTION_KEY=$(openssl rand -hex 32)" >> .env
echo "POSTGRES_PASSWORD=$(openssl rand -hex 16)" >> .env

docker compose up -d

Access n8n at http://localhost:5678 and create your admin account.

Docker Compose (Standalone)¶

Minimal setup when you already have Ollama running:

# docker-compose.yml
services:
  n8n:
    image: n8nio/n8n:latest
    restart: unless-stopped
    ports:
      - "5678:5678"
    environment:
      - N8N_ENCRYPTION_KEY=${N8N_ENCRYPTION_KEY}
    volumes:
      - n8n-data:/home/node/.n8n

volumes:
  n8n-data:

This uses the built-in SQLite database. Suitable for personal use and experimentation.

Configuration¶

Environment Variables¶

Variable	Description	Default
`N8N_ENCRYPTION_KEY`	Encrypts credentials at rest (required)	--
`DB_TYPE`	Database backend (`sqlite`, `postgresdb`)	`sqlite`
`DB_POSTGRESDB_HOST`	PostgreSQL host	`localhost`
`DB_POSTGRESDB_DATABASE`	Database name	`n8n`
`N8N_DIAGNOSTICS_ENABLED`	Send usage telemetry	`true`
`WEBHOOK_URL`	External webhook URL	auto-detected
`N8N_PORT`	HTTP port	`5678`

Ollama Credentials¶

Go to Settings > Credentials > Add Credential
Search for Ollama
Set Base URL:
- Docker (same Compose network): http://ollama:11434
- Docker (host Ollama): http://host.docker.internal:11434
- Native install: http://localhost:11434

AI Workflows¶

AI Agent Node¶

The AI Agent node connects an LLM to tools, memory, and output parsers:

Add an AI Agent node to your workflow
Connect an Ollama Chat Model sub-node:
- Select your Ollama credential
- Choose a model (e.g., llama3.3, qwen2.5)
Add tools (optional):
- HTTP Request -- call external APIs
- Code -- execute JavaScript/Python
- Postgres / MySQL -- query databases
- Vector Store -- search Qdrant/Chroma for RAG
Add memory (optional):
- Window Buffer Memory -- last N messages
- Qdrant Vector Store Memory -- long-term recall
Set a System Message to define the agent's behavior

Text Classification¶

Categorize input data using an LLM:

Add a Text Classifier node
Connect an Ollama Chat Model sub-node
Define categories (e.g., support, billing, feature-request)
The node outputs the matched category for downstream routing

Example use cases:

Classify incoming support tickets and route to the right team
Categorize emails before processing
Tag incoming webhook data

RAG with Qdrant¶

Document Ingestion Workflow¶

Build a workflow to load, chunk, embed, and store documents:

Read Binary File or HTTP Request -- load document
Extract Document Text -- extract text from PDF/DOCX
Recursive Character Text Splitter -- chunk text
Embeddings Ollama -- generate embeddings (nomic-embed-text)
Qdrant Vector Store (Insert) -- store chunks with embeddings

Query Workflow¶

Answer questions using stored documents:

Webhook or Chat Trigger -- receive question
AI Agent node with:
- Ollama Chat Model as LLM
- Vector Store Tool pointing to Qdrant collection
Respond to Webhook -- return the answer

Example: Document Q&A Bot¶

Step-by-step workflow:

Create a new workflow
Add a Chat Trigger node (provides built-in chat UI)
Add an AI Agent node and connect it to the trigger
Under the AI Agent, add:
- Ollama Chat Model -- select llama3.3
- Vector Store Tool -- connect to your Qdrant collection
- Window Buffer Memory -- keep last 10 messages

Set the system message:

You are a helpful assistant that answers questions based on the provided documents.
Always cite which document your answer comes from.
If you don't know the answer, say so.

Click Chat to test in the built-in chat panel

Troubleshooting¶

Ollama connection issues in Docker¶

If n8n cannot reach Ollama:

# Verify Ollama is accessible from the n8n container
docker exec -it n8n-n8n-1 wget -qO- http://host.docker.internal:11434/api/tags

If using separate Compose files, create a shared Docker network:

docker network create ai-network

Then add to both Compose files:

networks:
  default:
    external: true
    name: ai-network

Timeout on large model responses¶

For models that generate long responses, increase the HTTP timeout:

In the Ollama credential settings, set a longer timeout
Or set OLLAMA_KEEP_ALIVE to a longer duration on the Ollama side

Workflow execution errors¶

Check the execution log: Executions tab in the left sidebar
Each node shows its input/output data -- click a failed node to inspect
Enable Error Trigger node to catch and handle failures