AI Assistant on self-hosted n8n: easier setup in 2.35

Announcements
1

Hey community! :waving_hand:

We’re making the AI Assistant much easier to run on self-hosted n8n.

When we first shared early setup instructions, getting AI Assistant working on self-hosted required enabling the module, wiring up a sandbox service, configuring model settings, and optionally adding web search - mostly through environment variables.

With n8n 2.35, that setup is becoming much simpler.

A heads-up before you dive in: AI Assistant on self-hosted is still in preview. It can make mistakes, and its behavior may change. But the path from “I want to try this on my own instance” to “I’m building workflows with AI Assistant” should now be much faster.

You can find full details at https://docs.n8n.io/deploy/host-n8n/install-options/one-line-setup

What’s changed

The big change is that you no longer need to piece everything together manually from scratch.

For new Docker-based installs, there’s now a one-line setup path that installs n8n with the required sandbox and web search preconfigured:

curl -fsSL <https://get.n8n.io> | sh

This installs n8n with Docker Compose and pre-configures:

  • A code sandbox
  • SearXNG web search
  • AI Assistant enabled by default

You’ll still need to connect a model provider, but that part can now be done in the UI.

What you’ll need

To use AI Assistant on self-hosted n8n, you’ll need:

  • n8n 2.35 or later
  • A Docker-based n8n install
  • An API key from a supported model provider, such as Anthropic, OpenAI, OpenRouter, or another OpenAI-compatible endpoint
  • A sandbox provider
  • Optional, but recommended: a web search provider, such as SearXNG or Brave Search

n8n does not provide the model. You bring your own API key, and model usage is billed directly by your provider.

New installs

For a new self-hosted instance, the quickest way to get started is:

curl -fsSL <https://get.n8n.io> | sh

This installs n8n with Docker Compose and sets up the sandbox and SearXNG web search for you.

After that:

  1. Open n8n.
  2. Connect a model provider in the UI.
  3. Start building with AI Assistant.

The model provider is the only thing you need to add yourself.

Existing Docker installs

If you already run n8n with Docker:

  1. Update to n8n 2.35 or later.
  2. Follow the setup docs to configure the sandbox and, optionally, web search.
  3. Connect your model provider in the UI.

The AI Assistant module is enabled by default from 2.35.

You can use your own Docker-hosted sandbox if you want to keep everything on your infrastructure, or use a hosted sandbox provider such as Daytona.

Who this is for

This update is mainly for:

  • Self-hosted admins who want to enable AI Assistant on an existing Docker install
  • New users spinning up a self-hosted n8n instance
  • Teams that self-host for data control and want to choose their own model provider, sandbox, and search setup
  • Builders who want to get from an automation idea to a working workflow faster than building node by node

Where to post questions

If you have questions or feedback about self-hosted setup, please post them in this thread.

If you have feedback about what AI Assistant builds, how it behaves, or the workflow-building experience itself, please use the main AI Assistant announcement thread so we can keep product feedback in one place.

Happy automating!

6 Likes
Why are there two Docker Compose doc pages?
Assistant update: agents, MCP servers, one-line self-hosting, and more
Can't connect n8n Agents
2

Hi @PaulGordon

I am using the n8n Cloud plan.

I want an option to disable autosave and use the old manual Save behavior again.

Right now, whenever I make any change in a workflow, it gets saved automatically. For example, sometimes I only want to test something by adding a node, removing a node, changing a connection, or changing some settings. If I decide that I do not want those changes, I would prefer to simply leave the workflow without saving them.

Earlier, the manual Save button gave me better control because only the changes that I intentionally saved became part of the workflow.

I understand that self-hosted users can use:

N8N_WORKFLOWS_AUTOSAVE_DISABLED=true

So my request is to please add a setting for n8n Cloud users like:

Workflow Settings → Autosave

  • Autosave ON
  • Autosave OFF / Manual Save

When Manual Save is enabled, I would like the workflow to behave like before:

  • Changes should not be saved automatically
  • A Save button should be available
  • Ctrl + S / Cmd + S should save the workflow
  • If I leave the workflow without saving, my changes should be discarded

This would be very useful for users who frequently experiment with workflows and want more control before saving changes.

Please consider making this option available for n8n Cloud users as well.

Thank you.

1 Like
3

Remember to upvote your feature request as well

1 Like
4

A lot of production systems use Docker in rootless mode, and privileged containers don’t work in that mode. It would be good to have some docs on how to run it on rootless Docker in Docker.

I want to try running the sandbox in its own isolated VM to reduce risks. Running a privileged container in prod is too risky for me. I’ve been figuring that out in this thread: Why are there two Docker Compose doc pages? - #14 by Daniel . It looks like I can split sandbox-api and sandbox-runner-1 into a separate compose.yml that runs in a separate VM, then just adjust the URL n8n uses to communicate with the API.

The set up AI assistant docs say to use N8N_INSTANCE_AI_SANDBOX_API_URL to specify the sandbox-api URL, but I don’t see this used anywhere in the code. It seems to be called N8N_SANDBOX_SERVICE_URL now? Is that just because the page hasn’t been updated to reflect the changes this post is talking about? I was going to try and set this up using the code (rather than the docs) for guidance, but any pointers would be appreciated!

5

This sounds like podman

1 Like
6

I don’t have a lot of experience with podman, but that’s my understanding too. Docker has a rootless mode too though: Rootless mode | Docker Docs

1 Like
7

My question is about using local ollama as a provider after much trouble setting up the compose file so n8n could talk to ollama out of 6 models that should work none of them respond properly usually with errors or code output instead of chat text I am wondering if you have tested models that will work well and what context window I should use with a rtx 3090

8

Well thats me XD gonna try this bad boy
thanks <3