n8n has been the self-hosted automation tool of choice for teams who want Zapier-style workflow power without the per-task pricing or the data leaving their infrastructure. When version 2.0 landed in January 2026, it changed the platform’s centre of gravity pretty significantly. AI agents went from an afterthought you could bolt on with HTTP request nodes to a first-class part of the platform.

If you haven’t revisited n8n since before 2.0, or you’re evaluating it now for the first time and wondering whether it’s actually useful for AI agent workflows versus just automations, here’s what the landscape looks like mid-2026.

What 2.0 Actually Added

The headline change is native LangChain integration, surfaced through 70-plus dedicated AI nodes in the workflow canvas. Before 2.0, if you wanted to build something like a retrieval-augmented generation pipeline, you were doing it through custom HTTP request nodes calling LangChain’s REST API, or writing JavaScript function nodes to handle prompt chaining. It worked but it was fiddly, and every new model or provider you wanted to add meant maintaining more custom code.

In 2.0, LangChain is treated as a first-class component. You can drag a Chat Model node directly onto the canvas, connect it to a Memory node that stores conversation history in Redis or Postgres, hook up a tool node that calls another workflow as a sub-agent, and have the whole thing running in a few minutes without writing code. The agent builder lets you define what tools the agent has access to, what memory type it uses, and what the system prompt looks like, all through the visual interface.

The sub-workflow-as-tool pattern is particularly useful. You can take any existing n8n workflow — a workflow that searches your internal knowledge base, or calls your CRM, or sends a Slack message — and expose it to an AI agent as a callable tool. The agent can then decide, based on user input, which tool to invoke and when. This is the model that a lot of production agentic systems use, and having it available in n8n without any custom infrastructure is a meaningful step up from where things were a year ago.

Memory: The Part That Actually Matters

One of the consistent frustrations with early no-code AI integrations was that every conversation started from scratch. The AI had no memory of what you’d discussed before, which made it useless for anything except one-shot question answering.

n8n 2.0’s memory integration changes this properly. You can configure conversation memory to persist in:

  • In-process memory: ephemeral, only lasts for the session, fine for short interactions
  • Redis: fast, appropriate for active sessions where you need sub-second retrieval
  • Postgres: durable, good for long-running workflows where conversation history needs to survive across sessions or system restarts

For enterprise deployments where you’re building AI assistants that employees return to day after day, the Postgres-backed memory is the right choice. You’re keeping the conversation history in your own database infrastructure, which matters if you have data residency or compliance requirements.

What Works Well in Production

The use cases where n8n 2.0 AI agent workflows shine are fairly predictable: internal knowledge base assistants, customer support triage bots, and document processing pipelines.

The internal knowledge base assistant pattern is particularly well-suited to n8n because you typically already have your knowledge sources connected in n8n, your Notion databases, Confluence spaces, Google Docs, internal APIs. Adding an AI agent on top of those connections is natural rather than requiring you to rebuild the data access layer from scratch.

For teams on the technical side, the combination of n8n’s webhook support and its AI agent nodes makes it reasonably straightforward to build an agent that can receive requests from Slack, process them with an LLM, call internal tools, and respond back to Slack, all configured visually with minimal custom code.

The one area where you’ll still hit limits is complex multi-agent orchestration. n8n 2.0’s agent builder handles single-agent-with-tools well. If you’re building something where multiple specialist agents need to collaborate, hand off tasks to each other, and resolve conflicts, you’re likely to find n8n’s visual interface starts to feel constraining and you’ll be reaching for code nodes more frequently than you’d like.

Self-Hosted vs Cloud

n8n is available both self-hosted (Apache 2 licence for most features, a separate enterprise licence for some advanced features) and as a cloud service. For most teams building AI agent workflows, the self-hosted path is worth considering, especially if:

  • You’re processing sensitive data that can’t leave your infrastructure
  • You want to connect to internal services that aren’t internet-accessible
  • You’re using locally hosted models via Ollama or vLLM rather than cloud APIs
  • You want cost certainty rather than per-execution pricing

The self-hosted setup has gotten easier, a single Docker Compose file gets you running in minutes, and the 2.0 AI features are fully available in the community edition without requiring an enterprise licence.

Fair enough, if your team doesn’t want to maintain infrastructure, the cloud option is solid and the pricing is reasonable for moderate workflow volumes.

Where to Start

If you’re new to n8n and want to understand the AI agent capabilities, the official n8n AI agent templates are a good starting point. There are working examples for chat assistants, knowledge base Q&A, and automated research workflows that you can import and modify rather than building from scratch.

If you’re migrating from another automation platform or from a custom-built LangChain setup, the main question to answer first is whether n8n’s visual interface matches how your team thinks about workflows. Some engineering teams prefer code-first frameworks like LangGraph or AutoGen for the control they offer. n8n’s strength is speed of iteration and accessibility to non-engineers, not maximum control over every aspect of agent behaviour.

n8n 2.0 is a genuinely useful platform for a meaningful class of AI agent workflows. It’s not trying to compete with custom code for complex multi-agent systems, but for the practical majority of business automation use cases that need LLM capabilities, it’s a strong option.