n8n is an open-source workflow automation platform: the n8n repository maintained by the n8n-io organization on GitHub has passed 200,000 stars, making it one of the hottest automation infrastructure projects today. Think of it as "the open-source Zapier" — except you can install the entire platform on your own server: data never leaves your network, and self-hosting comes with no execution limits.
What exactly is n8n
In one sentence: a workflow automation platform combining a visual canvas with code nodes, with native AI agent orchestration.
Drag nodes onto the canvas to build a workflow: scheduled triggers, webhooks, API calls, database reads and writes; complex logic goes straight into Code nodes written in JavaScript or Python, with npm packages supported. In recent years n8n has pivoted toward AI: LangChain-based AI agent nodes let workflows call large language models directly, attach tools, and run multi-step reasoning, and the official template library already holds 9,000+ ready-made workflows. "n8n" is short for "nodemation", pronounced "n-eight-n".
Why it blew up
First, AI agents need an orchestration layer in production: models do the thinking, but tool calls, branching, and human approvals do the actual work — agent orchestration approaches like LangGraph solve the same class of problem, while n8n gives non-coding teams a visual interface. Second, data sovereignty: finance and healthcare scenarios can't hand data to a third-party SaaS. Third, pricing: the official cloud charges per execution, while self-hosting only costs server money — high-volume teams do the math and switch.
Who it's for — and who should skip it
Good fit: indie developers and small teams with basic ops skills; companies whose data can't leave the intranet; teams wiring AI agents into existing systems with frequent internal API and database calls; high-volume teams sensitive to per-execution SaaS billing.
Not for: anyone who never wants to touch a server — self-hosting means you own updates, backups, and security; teams that only need two or three simple automations and are fully non-technical will have an easier time with Zapier or Make's pure SaaS experience. One more note: n8n uses a fair-code license (Sustainable Use License) — source-available but not OSI-approved open source. If your legal team is strict about license types, review the license text before adopting.
Deployment cost: the part measured in real money
Getting started is fast: one Docker command gets it running, SQLite by default, open port 5678 in the browser and you're in — that's the "trial cost".
Production is a different ledger: light trials start at 1 vCPU and 2GB RAM; real workloads call for 2 vCPUs and 4GB plus PostgreSQL; heavy concurrency or long-running AI executions need queue mode (Redis as the task queue, main process separated from workers). A VPS that meets the bar costs roughly $4–10/month, versus the official cloud starter at €24/month with execution caps — the hardware cost is genuinely lower. To run models locally too, see this cost analysis of self-hosting Ollama. Don't forget the hidden costs: security patches, version upgrades, backups, and 3am alert wake-ups are ongoing labor.
Four things to know before self-hosting
1. Lose the encryption key, lose every stored credential. n8n encrypts all third-party credentials with N8N_ENCRYPTION_KEY, stored in the /home/node/.n8n directory. The official docs warn: if the directory is missing at startup, n8n generates a new key automatically and old credentials can never be decrypted again. Community members have lost the key migrating from single-instance to queue mode and watched every connection die overnight. Countermeasure: set the key explicitly on day one, back it up off-server, and migrate the key before the data.
2. Breaking changes in major upgrades are routine. n8n ships fast and majors regularly break things: 2.0 moved the task runner out of the main image, so self-hosters need the separate n8nio/runners image; 3.0 (planned October 2026) will require Docker for self-hosting, drop npm installs, and remove a batch of legacy nodes. Database migrations only move forward — there is no official rollback path. Countermeasure: pin the image version in production, back up the database before upgrading, and scan affected workflows with the official migration report first.
3. The default setup can't handle concurrency and heavy loads. Default SQLite plus single-process: once concurrency rises or a single workflow carries large payloads, the most common community complaint is memory ballooning or container crashes. Production needs PostgreSQL, automatic execution pruning, and splitting into sub-workflows when memory gets tight. AI agent users beware: long executions and multi-round tool calls eat far more memory than traditional automation — plan 1–2GB per worker in queue mode.
4. Self-hosting means owning the entire security burden. A September 2026 GitGuardian study scanning public Git commits found 4,576 leaked n8n API tokens; another 129 public instances were still running on previously leaked weak encryption keys. Exposed ports, reverse proxies, HTTPS, patching — everything the platform did for you in the SaaS era is now yours. Countermeasure: don't expose port 5678 publicly, manage tokens and keys via environment variables and a secrets manager, and require HTTPS in production.
Official repository info
- Platform: GitHub
- Organization: n8n-io
- Project: n8n (repository n8n-io/n8n)
- License: Sustainable Use License (fair-code: source-available, self-hostable; not OSI-approved open source)
