The n8n team keeps up a relentless release cadence — a new minor most weeks. The 2.40 line is the current stable series, and it brings one feature AI-workflow builders have been asking for.
What’s new in 2.40
- Fallback models on AI nodes — you can now configure a secondary model alongside the primary one. If the main provider errors or times out, the node retries on the fallback instead of failing the workflow.
- AI agent fixes — the agent’s “thinking” blocks render correctly again, and credential handling around AI nodes got tightened up.
- Assorted fixes — including Confluence page ordering in the Confluence node, gateway credit eligibility checks on sub-nodes, and a warning for the deprecated
N8N_DB_PING_TIMEOUTvariable ahead of v3 storage changes.
Also worth knowing from the 2.x series
If you haven’t looked at n8n since 1.x, a lot has changed:
- AI Assistant — describe a goal in plain language and get a working draft workflow on the canvas.
- Canvas Groups — organize large workflows into labeled sections instead of a spaghetti of nodes.
- Verified webhooks — fourteen trigger nodes can now verify webhook signatures, which matters once your instance faces the internet.
- Webhook responses of any size — workers can stream back large responses instead of hitting payload limits.
- New model providers — Moonshot Kimi and Alibaba Cloud Model Studio joined the AI node roster.
- Workflow packages — move workflows between instances as packaged exports.
- OpenTelemetry from the UI — configure tracing without editing environment files.
How to try it
Self-hosting n8n the classic way means a server, Docker, a database, HTTPS, and then keeping it updated. The point of trying new features quickly is that you shouldn’t need a lab weekend for it.
The Pod card below runs a dedicated n8n instance for you — embedded SQLite, persistent storage for your workflows and credentials, and its own HTTPS Address. Create the Pod, open n8n, and 2.40 is what you get.