Can you build AI agents in n8n?

What n8n Provides for AI Agents

n8n is a fair-code workflow engine, founded in 2019 in Berlin, that pairs a visual editor with code blocks. The AI Agent node, generally available since 2024 and significantly extended through 2025-2026, wraps LangChain primitives so a workflow can reason, call tools, and persist memory inside the regular n8n canvas.

Capabilities

The AI Agent node accepts:

• A chat model (OpenAI, Anthropic, Mistral, Google Vertex AI, Ollama, or any OpenAI-compatible endpoint)
• A list of tools (other n8n sub-workflows or HTTP requests)
• An optional memory backend (Window Buffer, Summary, or Postgres-backed)
• An optional vector store as a retriever (pgvector, Pinecone, Weaviate, Qdrant, Milvus)

The node runs a ReAct or function-calling loop until the model emits a stop condition or a configurable max-iteration cap is reached.

Common Patterns

Production agent patterns built in n8n include:

• Support triage: read a Zendesk ticket, classify, retrieve KB articles, draft a reply
• Sales research: search the web, read pages, summarise into Notion or Salesforce
• Internal copilot: query Postgres, answer a Slack question with citations

Deployment

The AI Agent node is available on n8n Cloud (Starter $24/month, Pro $60/month, Enterprise custom as of May 2026) and on the free, self-hosted Community Edition. For workloads with long-running model calls or local model usage via Ollama, self-hosting on a Linux host or Kubernetes namespace is the typical choice.

Caveats

n8n is well suited to agents embedded in business workflows (CRM updates, ticket routing). For standalone consumer chat surfaces with rich conversation UIs, code-first frameworks like LangGraph or CrewAI plus a dedicated frontend offer more control over the conversation loop.