The official n8n desktop app is no longer maintained: its GitHub repository was archived in August 2025. This guide covers the practical replacement on Windows, which is running self-hosted n8n locally through Docker Desktop and using the editor in your browser. Docker Desktop is not a native n8n desktop application. It’s the container runtime that keeps n8n running in the background. You’ll install it, start n8n, build a test workflow, restart everything, and confirm your work survives.
Documentation checked October 28, 2026. This guide has not been tested hands-on; the commands and expected outputs should be verified on your Windows setup before you rely on it for production.
If you are still choosing where n8n should run, start with our self-hosted n8n vs n8n Cloud comparison. This guide focuses on a local Windows setup for learning and testing.
1. Can You Still Download the n8n Desktop App?
Not from n8n. The official n8n-desktop-app repository was archived by its owner on August 15, 2025, which makes it read-only, and its README says only that the repository is no longer maintained. There’s no newer official desktop release behind it.
That explains a question newcomers keep running into. In a September 2025 community thread, someone new to n8n asked how to download the desktop version because several YouTube tutorials used it. One reply pointed to a Docker guide and an npm guide. Another said the desktop app is no longer officially maintained and suggested running npx n8n. The thread doesn’t name the videos, so the following is my interpretation: a tutorial that shows a desktop installer is describing software n8n has stopped maintaining, and the screen you’re hunting for won’t appear.
Search results for “n8n desktop download” also surface independent wrappers and forks. The archived repository alone has 112 forks. I haven’t evaluated any of them, and this guide doesn’t use them. An installer that bundles n8n is someone else’s build, and you’d be trusting it with every credential you add later.
The software used here comes from two places: Docker Desktop from Docker’s Windows install page, and the n8n image from n8n’s own registry, docker.n8n.io/n8nio/n8n. If you’d rather not run anything yourself, n8n Cloud is the official hosted route (see the FAQ).

What you’ll end up with: a Windows machine running Docker Desktop’s Linux engine (the WSL 2 backend), an n8n container inside it, and a browser tab at localhost:5678. The browser is only the interface. The n8n process runs on your computer, and if the container stops, the editor goes with it.
2. Install n8n on Windows with Docker Desktop
Every command below is PowerShell, in Windows Terminal or the PowerShell app. I’ve kept to one environment on purpose.
n8n’s Docker Compose guide tells Windows users to work inside WSL and keep project folders on the WSL filesystem, because bind mounts across /mnt/c/... are slow and can cause permission issues. That warning is about bind-mounted folders. This setup stores n8n’s data in a Docker-managed volume instead, so the only file on the Windows side is the Compose file.
One more thing to know before you start. That same guide now builds a larger stack that includes n8n Assistant, a privileged sandbox runner, and a search service, and it asks for at least 4 GB of RAM and 2 vCPUs. The file below is a reduced version I assembled from n8n’s documented settings, not an official minimal example. It leaves the Assistant out because you don’t need it for a first workflow.
Check the prerequisites
Docker’s page lists these for the WSL 2 backend: Windows 10 22H2 (build 19045) or Windows 11 23H2 (build 22631) or newer, the WSL 2 feature turned on, WSL version 2.1.5 or later, 8 GB of RAM, and hardware virtualization enabled in BIOS/UEFI. Check WSL first:
wsl --version
If that prints version details and the number is 2.1.5 or higher, you’re set. If it prints nothing useful, you’re probably on the old inbox WSL. Run wsl --update in an administrator terminal. Enabling WSL 2 for the first time needs admin rights; Docker Desktop’s default per-user install doesn’t.
Docker Desktop is free for personal use, education, and small businesses (fewer than 250 employees and under $10 million in annual revenue). Larger companies need a paid subscription, which matters if you’re setting this up for work.
Install it from the page above, then start Docker Desktop from the Start menu. It doesn’t launch by itself after installation, and you’ll need to accept the terms before the engine starts.
Check the engine and Compose
docker versiondocker compose version
Success means docker version prints both a Client section and a Server section, and docker compose version prints a v2 version number. If you see Client details followed by a connection error, the engine isn’t running yet; jump to the troubleshooting section.
Create the project folder and Compose file
mkdir $HOME\n8n-localcd $HOME\n8n-local
Now write the configuration. This here-string avoids Notepad’s habit of saving compose.yaml as compose.yaml.txt:
@'name: n8n-localservices: n8n: image: docker.n8n.io/n8nio/n8n:2.45.4 container_name: n8n restart: unless-stopped ports: - "127.0.0.1:5678:5678" environment: - GENERIC_TIMEZONE=Europe/Berlin - TZ=Europe/Berlin - N8N_ENFORCE_SETTINGS_FILE_PERMISSIONS=true volumes: - n8n_data:/home/node/.n8nvolumes: n8n_data: name: n8n_local_data'@ | Set-Content -Path compose.yaml -Encoding ascii
Change Europe/Berlin to your own timezone name in both places. Here’s what each part does:
- Image version.
2.45.4was the newest stable release on n8n’s releases page on October 28, 2026. A 3.0.0 pre-release had appeared the day before, and I’d stay on stable. Pinning matters because an unpinned image changes whenever you pull, and updates can include database migrations. You decide when the version changes. - Port binding.
127.0.0.1:5678:5678maps your computer’s port 5678, on the loopback address only, to the container’s 5678. The editor stays off your local network. Drop the127.0.0.1:prefix and Docker publishes the port on all interfaces, which is what the plain5678:5678in n8n’s examples does. - Persistent storage. The named volume is mounted at
/home/node/.n8n. According to the backup documentation, that folder holds the config file with the credential encryption key and, with the default SQLite database, the database itself. I gave the volume an explicit name so the backup commands later don’t depend on a folder-based prefix. - Timezone.
GENERIC_TIMEZONEsets the zone used by schedule-oriented nodes.TZsets the container’s system timezone. Without the first, schedules run in whatever default n8n picks. - File permissions.
N8N_ENFORCE_SETTINGS_FILE_PERMISSIONScomes from n8n’s official Docker example and enforces secure permissions on the settings file. - Restart policy.
unless-stoppedbrings the container back when Docker Desktop starts. It can’t run while Windows is asleep or off.
Two omissions are deliberate. There’s no database service, so n8n falls back to SQLite, which n8n’s docs consider fine for trying things out; they recommend PostgreSQL for production. And N8N_RUNNERS_ENABLED, which older tutorials include, is deprecated from n8n 2.0 and no longer needed.
In short, this configuration gives you one n8n container, one SQLite database inside one named volume, an editor reachable only from your own PC, and a pinned version. It does not give you webhooks that outside services can reach.

Start n8n
docker compose up -ddocker compose psdocker compose logs --tail 30 n8n
The first run downloads the image, so give it a few minutes. A successful result looks like this: docker compose ps lists n8n as running with 127.0.0.1:5678->5678/tcp, and the last log lines say the editor is accessible at http://localhost:5678. For a direct check, use curl.exe (plain curl is an alias for something else in Windows PowerShell):
curl.exe -s http://127.0.0.1:5678/healthz
It should return a short JSON status saying the instance is okay.
Open the editor and create the owner account
Go to http://localhost:5678. n8n shows an owner setup form. The account lives only in your local database; it isn’t an n8n Cloud login. Use a password you’ll keep in a password manager, because a local instance has no “forgot password” email unless you configure one. Optional onboarding questions may follow, and you can skip them. You’re done when you land on an empty workflow list.
3. Build a Test Workflow Before Connecting Your Accounts
A workflow with no credentials separates infrastructure problems from account problems. If something breaks later, you’ll know this part worked.
Once a workflow starts, a failure inside it is a different problem from an unreachable editor. Our n8n error workflow guide explains why failure alerts may not arrive.

- Create a new workflow and add Trigger manually as the first node.
- Add an Edit Fields (Set) node after it.
- In Edit Fields, add two fields in manual mapping mode:
messagewith the valuen8n is running locally, andsetupwith the valueWindows + Docker Desktop. - Select Execute workflow.
- Open the Edit Fields node and look at its Output panel.
- Rename the workflow to
Local setup checkand save it with Ctrl+S.
Success is one output item containing exactly those two fields. If the node shows no output, the run didn’t reach it; check that the trigger is connected.
One version note. Since n8n 2.0, saving stores a draft, and the old Active toggle became Publish and Unpublish. A manual workflow never needs publishing, since you run it yourself. Triggers that fire on their own, like webhooks and schedules, only run from the published version.
4. Restart n8n and Check That Your Workflow Is Still There
docker compose restart n8n
Wait a few seconds, reload the browser, sign in, and open Local setup check. Run it again; the output should match what you saw before.
That restart keeps the same container, which is a weak test. The stronger one replaces the container entirely:
docker compose downdocker compose up -d
This removes the container and the project network, then builds a fresh container from your Compose file. The named volume stays. Sign in again; your account and workflow should both be there. If they are, persistence works.

The four things people mix up behave quite differently:
- Closing the browser. Nothing happens to n8n. The tab is only a client.
- Stopping the container (
docker compose stop, or quitting Docker Desktop). The n8n process stops, but the container and its storage remain. Nothing is reachable or running until you start it again. - Removing the container (
docker compose down). The container disappears and the volume doesn’t. Recreating it fromcompose.yamlbrings everything back. - Removing the volume. This is the only one that deletes your database and your encryption key. Docker Desktop’s Volumes tab can do it, and so can adding the
-vflag todown. There’s no routine reason to do either.
5. Fix Common n8n Setup Problems on Windows
Docker can’t connect to the engine
First, confirm Docker Desktop is actually running and has finished starting. Then run docker version. If the Client section appears but the Server section is replaced by an error, usually one that mentions a dockerDesktopLinuxEngine pipe, the engine isn’t up.
Next, run wsl --version. No output means the old inbox WSL, and wsl --update from an administrator terminal fixes that. If WSL looks fine, check that virtualization is enabled in BIOS/UEFI, which Docker’s system requirements list as a prerequisite.
localhost:5678 doesn’t open
Start with docker compose ps. If n8n isn’t running, read docker compose logs --tail 50 n8n. If it is running, run the curl.exe health check from earlier. When that works but the browser doesn’t, try http://127.0.0.1:5678 directly, then look at browser extensions or a proxy setting. After a version change, give n8n a minute: it may be migrating its database, and the logs will show it.
Port 5678 is already taken
Docker will refuse to start the container. The first check is docker ps, because the usual culprit is an earlier test container still holding the port. If it’s not that, find the process:
Get-NetTCPConnection -LocalPort 5678 -State Listen | Select-Object LocalAddress, OwningProcessGet-Process -Id <the OwningProcess number>
Stop whatever you find if you don’t need it. Otherwise change only the left-hand number in ports, for example "127.0.0.1:5679:5678", and open localhost:5679. Leave the right-hand number alone; it’s the container’s internal port. A leftover container that’s also named n8n will cause a separate name-conflict error; stop and remove it only after you’ve confirmed it holds nothing you need.
n8n shows a fresh setup screen
The first check is what’s mounted:
docker inspect n8n --format '{{ json .Mounts }}'docker volume ls
You should see the n8n_local_data volume mounted at /home/node/.n8n. If the mount is missing or points elsewhere, n8n created a blank database and is showing you a new installation. Your old data is probably still sitting in whichever volume or folder it was written to.
A documented case: in a September 2025 community thread, a Docker Desktop user pulled a newer n8n image and could no longer log in. A respondent suspected the container was replaced without persisted data. The poster replied that they still had an n8n-data folder containing a database file of about 6 MB, and the follow-up advice was to confirm that folder was mounted at /home/node/.n8n. The thread doesn’t show how it ended, so I can’t say that fix worked. My reading is that it’s the same symptom as above, which is why checking the mount comes first.
A workflow stops when the laptop sleeps
Open the Executions list and look for a gap that lines up with when the computer slept or was shut. When Windows sleeps, the Docker engine isn’t doing any work, so nothing runs. I wouldn’t assume that runs missed during sleep get replayed.
Here’s a hypothetical example: a lead-notification workflow that polls a form every few minutes. The owner closes the laptop at 6 p.m. and the workflow goes quiet until morning, so leads from the evening reach the sales team late. Changing Windows power settings can help on a desk machine. For anything that has to run overnight, you need an always-on host.
An external service can’t reach a local webhook
First, look at the URL your Webhook node displays. It’s built on localhost, and a service on the internet can’t call your localhost. A private address like 192.168.x.x doesn’t fix that, since it’s only meaningful inside your network. n8n’s npm install page says it directly: for trigger nodes of external services such as GitHub, n8n has to be reachable from the web.
Two more checks. Make sure the provider is using the production URL, not the test one. In n8n 2.x the production URL only responds once the workflow is published. For short experiments, n8n documents a tunnel built on cloudflared, but it says plainly that the tunnel is for development and testing and isn’t safe for production. For real use, run n8n on an always-on machine with a public HTTPS address and set its WEBHOOK_URL to match. Opening ports on your home router is not where I’d start.
For a Telegram bot that sends messages but never receives them, follow our n8n Telegram Trigger troubleshooting guide to check webhook registration, public HTTPS access and competing bot connections.

6. Back Up Your Local n8n Before Updating It
Two different things get called a backup. Exporting workflows writes them to JSON files, which is handy for moving workflows around. Per the backup documentation, those exports hold workflows and credentials only, and leave out users, execution history, variables, and instance settings, including the encryption key. Backing up the instance means saving the .n8n folder, which for this setup is the whole volume.
The encryption key matters most. n8n stores credentials encrypted, and without the key from the config file (or a custom N8N_ENCRYPTION_KEY) a restored database can’t be decrypted. That also means a backup archive containing the key is sensitive. Store it like a password vault.
The docs say to stop n8n before copying when you use SQLite, because copying a database mid-write can produce an inconsistent file. This procedure does that, and writes the archive outside the container:
cd $HOME\n8n-localNew-Item -ItemType Directory -Force -Path backups | Out-Nulldocker compose stop n8ndocker run --rm -v n8n_local_data:/source:ro -v "${PWD}\backups:/backup" alpine tar czf "/backup/n8n-data-$(Get-Date -Format yyyy-MM-dd).tar.gz" -C /source .docker compose start n8n
The temporary container reads the volume read-only and writes a compressed archive into backups in your project folder. It downloads a small Alpine image the first time. Check the result by listing the archive; it should include config and database.sqlite:
tar -tzf backups\n8n-data-2026-10-28.tar.gz
Substitute the file name from your own backups folder. A folder on the same disk isn’t much protection, so copy the archive to another drive or storage you trust. Practice a restore on a throwaway project before you need one, following the docs’ full-instance steps: stop n8n, restore the .n8n contents, restore your deployment configuration, start n8n.
When you’re ready to update, take a backup, change the tag in compose.yaml, then run docker compose pull followed by docker compose up -d. Don’t treat a tag change as an undo button. If a new version migrated the database, an older image may not understand it. n8n’s npm page notes that reverting after a migration can require n8n db:revert and a look at the release notes. The dependable way back is restoring the backup you took first.

7. Already Using the Old Desktop App?
Expect some care here. In a June 2023 community thread, a Windows 10 user on n8n 0.216.2 started n8n in Docker after the desktop app was deprecated, and none of the old workflows appeared. An n8n staff member suggested the documented Docker command uses a different folder than the desktop app, which they thought lived under C:\Users\.... The user then ran Docker with their C:\Users\<name>\.n8n folder mounted and reported that it worked, including starting active workflows. That’s one user’s report from 2023 on a very old version. Today’s releases are a long way from 0.216, so mounting an old data folder straight into the newest image is exactly what I’d avoid.
A safer sequence:
- Preserve the original. Close the desktop app and copy its data folder somewhere safe. The thread suggests it sits in your Windows user profile, but I haven’t confirmed the location for every installation.
- Check the old version. Open the desktop app and note its n8n version. The gap between that number and 2.x tells you how much has changed.
- Move workflows first. Export them from the old editor as JSON and import them into the new instance. Imported workflows arrive deactivated, per the backup docs.
- Handle credentials separately. The new instance can only decrypt them with the old encryption key, or from a decrypted export, which writes secrets in plain text and should be deleted afterwards. Re-entering credentials by hand is often simpler.
- Verify. Run each workflow manually, reconnect any credential that fails, and test triggers one at a time.
I haven’t verified a supported direct upgrade path from 0.x databases, which is why workflow export and import is the safer first route.
8. FAQ
Can I run n8n without Docker? Yes, with npm. n8n’s npm page says you need Node.js 20.19 to 24.x, and npx n8n starts it without a global install. It also says npm-based installs are deprecated from n8n 3.0, which has just appeared as a pre-release, and that n8n Assistant isn’t available with npm. For something you’ll rely on, Docker is the safer bet.
Does local n8n work offline? The editor and workflows made of local-only nodes should keep working once the image is on your machine. Anything that calls an outside API, plus the first image download, needs a connection.
Do I need an n8n Cloud subscription? No. The self-hosted edition used here is distributed under n8n’s Sustainable Use License. Cloud is the alternative if you’d rather not host anything.
Can I run it on a Mac? Yes. Docker Desktop has a Mac installer, and the Compose file should work unchanged. The PowerShell commands, especially the backup one, need Terminal equivalents.
Is a laptop suitable for client workflows? It’s good for building and testing. It’s a poor host for anything clients depend on, since it sleeps, moves between networks, has no public address for incoming webhooks, and is a single point of failure. Move those workflows to an always-on machine, with PostgreSQL as n8n’s docs recommend for production.
Before moving from a local prototype to a client deployment, use our n8n client handover checklist for deployment ownership, credentials, backups and acceptance testing.





Leave a Reply