What a self-hosted automation tool actually gives you
Self-hosted automation tools run the workflow engine on your own server, so executions and credentials stay in your database. n8n suits app-to-app workflows, Activepieces is MIT-licensed and simpler, Node-RED fits IoT and MQTT. Start with 2 vCPU and 4 GB RAM, Docker, PostgreSQL and a nightly backup; a VPS handles that.
The alternative is Zapier or Make, where you rent the engine per task. Self-hosting swaps a monthly task count for a fixed server bill and the work of running it. That trade pays off when you run thousands of executions a month, when workflows touch customer data, or when a connector you need does not exist.
- Executions are not metered. A polling workflow every five minutes costs nothing beyond CPU.
- Credentials live in your database, not a vendor's vault, so a DPDP review has one less party.
- You own upgrades, backups and uptime. Nobody else restarts the container at 2am.
The three tools worth running in 2026
Each of these runs on a Linux server with Docker and has a browser editor. They differ in licence, model and the jobs they are good at.
| Tool | Licence | Model | Best for |
|---|---|---|---|
| n8n | Fair-code (Sustainable Use Licence) | Node-based canvas, 400+ integrations, code nodes in JavaScript or Python | App-to-app workflows, CRM and invoice automation, teams leaving Zapier |
| Activepieces | MIT | Piece-based canvas, TypeScript, simpler editor | Small teams that want a genuinely open licence and fewer moving parts |
| Node-RED | Apache 2.0 | Flow-based, Node.js, strong MQTT and HTTP nodes | IoT, device data, internal tooling and simple scheduled jobs |
n8n is the strongest general choice and the one most teams migrate to. Activepieces is easier to read and has no usage restriction in its licence. Node-RED is older and narrower, but it is unbeatable for MQTT and cheap to run.
Server specs each tool actually needs
Vendor minimums assume a demo. These are working starting points for a small business running real workflows.
| Tool | vCPU / RAM | Disk | Database |
|---|---|---|---|
| n8n | 2 vCPU / 4 GB | 40 GB SSD | PostgreSQL from the start; SQLite is fine for testing only |
| Activepieces | 2 vCPU / 2 GB | 20 GB SSD | PostgreSQL, with Redis if you add workers |
| Node-RED | 1 vCPU / 1 GB | 10 GB SSD | SQLite or a file-based store is enough |
Two things drive the real number: how often workflows poll, and how much payload they move. A five-minute poll across ten services keeps a container awake all day, so give it memory rather than CPU. Keep execution history pruned; n8n stores every run and that table grows fastest.
Install n8n with Docker on a VPS
The Docker image is the supported install and avoids Node.js version drift on the host. Run these steps as a sudo-capable user on Ubuntu 22.04 or 24.04.
- Install Docker: curl -fsSL https://get.docker.com | sudo sh, then sudo usermod -aG docker $USER and log out and back in.
- Create a volume so workflows survive a container rebuild: docker volume create n8n_data.
- Start the container: docker run -d --name n8n --restart unless-stopped -p 5678:5678 -v n8n_data:/home/node/.n8n docker.n8n.io/n8nio/n8n.
- Open port 5678 to yourself only: sudo ufw allow from YOUR_IP to any port 5678, and do not open it to the world.
- Visit http://YOUR_SERVER_IP:5678 and create the owner account on the first screen. That account is local to your instance.
- Put a domain and TLS in front before you build production workflows, then set WEBHOOK_URL so webhook nodes print the public address.
You know it worked when the editor loads, the owner account signs in, and docker logs n8n shows the editor accessible line with no restart loop. If the page never loads, the port is closed at the firewall or the container exited; docker ps -a shows which.
Webhooks, TLS and backups on an automation server
A webhook URL is a public door into your workflows. Put the editor behind a domain with a certificate, keep 5678 closed to the internet, and treat any webhook path as an unauthenticated endpoint unless you add a header check.
Back up two things, not one: the database and the encryption key. n8n encrypts stored credentials with a key inside the data volume, so a database dump restored without that key leaves every credential unreadable. Copy the whole volume, or run the database dump and keep ~/.n8n/config together. Server backup solutions cover the off-server copy.
Test a restore before you need one. A backup nobody has restored is a guess, not a backup.
What this costs to run in India
The software is free to self-host; the server is the cost. A shared hosting plan will not run Docker or a long-lived Node process, so the floor is a VPS with root, or a cloud plan where the provider owns the hypervisor.
WebJi® sells both. VPS configurations are quoted by email before payment because RAM, disk and bandwidth move the price, and cloud plans are printed on the site with specs. Support is by email 24/7 in Hindi and English, and the office phone is business hours only. What we do not sell is a managed n8n service: you get the server, you run the stack.
When self-hosting automation is the wrong choice
Choose a hosted tool instead when nobody on the team will patch a container, when the volume is under a few hundred executions a month, or when the workflow is business-critical and needs a support contract behind it.
WhatsApp Business API, payment gateways and GST e-invoicing all have their own onboarding. Self-hosting the automation layer does not remove those steps, and a workflow that depends on an unofficial endpoint will break. Keep the critical path documented and the credentials in a password manager you control.
If the goal is a single scheduled job, cron on the same server is enough. Automation platforms earn their place when several systems have to talk to each other. For a comparable look at running another self-hosted workload, read self-hosted CCTV on a VPS.
Frequently Asked Questions
What are the best self-hosted automation tools?
n8n for app-to-app workflows, Activepieces for a simpler MIT-licensed editor, and Node-RED for IoT and MQTT flows. All three run in Docker on a Linux server and are edited in a browser. n8n has the widest integration list.
Is n8n free to self-host?
The n8n Community Edition is free to run on your own server under its fair-code licence, which restricts reselling it as a service. You pay for the VPS and your own maintenance. The paid cloud and enterprise tiers add features you do not get in the community image.
How much RAM does n8n need?
Plan for 4 GB on a 2 vCPU VPS for a small business workload running many polling workflows. 2 GB works for a handful of flows with little history. Memory, not CPU, is what runs out first once execution history grows.
Can I run n8n on shared hosting?
No. Shared hosting does not give you Docker, root or a long-running Node process, so the container cannot start. You need a VPS or a cloud plan with root access. A cPanel plan can host the website that triggers a workflow, not the workflow engine itself.
Is my data safe with self-hosted automation?
It stays on your server, which removes a third party from the data path but puts the security work on you: TLS, firewall rules, patched images and encrypted backups. Back up the encryption key with the database, or stored credentials cannot be restored.
Do self-hosted automation tools send email and WhatsApp messages?
Email works through SMTP or a transactional API once you add the credentials. WhatsApp needs an approved Business API provider and template messages, which applies whether the engine is hosted or self-hosted. Unofficial WhatsApp endpoints get accounts banned.