n8n is having a moment. It raised a $180M Series C at a $2.5B valuation in late 2025, took a strategic SAP investment in May 2026 that doubled its valuation to $5.2B, and sits at roughly 192,000 GitHub stars (as of June 2026). It is the default answer to “how do I automate this without paying Zapier.”
So why are so many people - maybe you, reading this - searching for n8n alternatives? It is almost never because n8n is bad.
Three honest reasons, and they are worth separating before you switch:
- License. n8n is source-available under the Sustainable Use License, not OSI-approved open source. This “fair-code” license lets you self-host and use it internally for free but forbids reselling it as a hosted service. That is fine for most people, but if you need a permissive (MIT/Apache) or OSI-approved license, n8n does not qualify.
- Resource appetite. n8n’s own docs recommend 2GB of RAM minimum and 4GB to be comfortable, and it spikes to 1-2GB on heavy executions. On a Raspberry Pi or a cheap VPS, lighter tools win.
- You would rather write code. n8n is visual-first. Plenty of developers want their automations as version-controlled code in Git, reviewed in pull requests, not as JSON node graphs.
Quick answer: The closest open-source n8n alternative is Activepieces (MIT-licensed, AI-native, no-code). If you would rather write code than drag nodes, look at Windmill (Python/TS/Go) or Trigger.dev (TypeScript). For lightweight or IoT use, Node-RED runs on a Raspberry Pi. For data pipelines, reach for Kestra or Apache Airflow. All nine are compared below by license, footprint, and use case.
And if you like n8n and only want it cheaper, self-hosting n8n runs $7/mo flat versus $24/mo on n8n Cloud. The alternatives below are for when you genuinely need a different tool.
Visual and No-Code (the closest swaps for n8n)
1. Activepieces - the open-source pick most people want
TL;DR: The closest truly open-source (MIT) swap for n8n - a no-code visual builder with ~280 AI/MCP-native connectors that runs on a 1.5-2GB VPS. Pick it if you want n8n’s feel without the fair-code license.
Activepieces is a no-code automation tool with a visual flow builder that bills itself as the “open source replacement for Zapier.” It is the closest match on the two things people leave n8n for: a real open-source license and a no-code visual builder.
License: MIT for the community edition - a genuine open-source license, unlike n8n’s Sustainable Use License. Enterprise features live in a separate commercial tier, but the core and the connectors are MIT.
Self-host: Docker Compose. It needs PostgreSQL and Redis, and the official production spec is heavy (multiple worker replicas), but a small instance runs fine on a 1.5-2GB VPS.
Where it wins: It is AI-native. Activepieces exposes its ~280 “pieces” (connectors) as MCP servers, so AI agents in Claude, Cursor, or Windsurf can call them directly - it markets itself as the largest open-source MCP toolkit. Connectors are type-safe TypeScript npm packages, and about 60% of them are community-built.
Watch out for: Fewer integrations than n8n (~280 pieces vs n8n’s 400+ nodes plus ~900 community nodes), and it is less battle-tested for deep, branching, developer-heavy workflows.
Best for: Anyone who wants a truly MIT-licensed, AI/MCP-native, no-code Zapier alternative and does not need n8n’s full long-tail connector catalog. ~22k stars, actively developed (the 0.78.x line shipped in early 2026).
2. Node-RED - the lightweight, permissive veteran
TL;DR: The lightest option here - Apache 2.0, idles around 100MB, runs on a Raspberry Pi. Best for IoT, edge, and real-time event wiring, not polished SaaS automation.
Node-RED is a flow-based programming tool: you wire “nodes” on a canvas and drop into JavaScript function nodes when you need custom logic. It started at IBM as an IoT and edge tool and is now an OpenJS Foundation project.
License: Apache 2.0. Fully permissive, vendor-neutral governance, no commercial gating.
Self-host: The easiest and lightest on this list. Install via npm or the official Docker image, default port 1880. A typical idle instance uses roughly 60-120MB of RAM and runs happily on a Raspberry Pi.
Where it wins: Resource footprint and ecosystem. There are 5,000+ community node packages and thousands of shared flows, and it is best-in-class for IoT, MQTT, hardware, and real-time event wiring.
Watch out for: It is not a polished SaaS-automation tool. Few turnkey business connectors ship in core (you assemble HTTP plus community nodes), data is passed as msg.payload objects, and there are no first-party AI agent nodes. It is engineer-oriented.
Best for: IoT, edge, home automation, and real-time event flows on a small box, where a permissive license and low RAM matter. ~23k stars, with version 5.0 in beta and v4.1.x shipping maintenance releases through 2026.
3. Automatisch - the simplest Zapier-style builder
TL;DR: A simple, AGPL-licensed Zapier-style builder for non-technical users. Easy to self-host, but only ~80 connectors and roughly one release a year - pick it for a few common automations, not power use.
Automatisch is an open-source tool that explicitly markets itself as “the open source Zapier alternative.” You build linear flows (a trigger, then a chain of actions) in a clean web UI with no code.
License: AGPL-3.0 for the community edition, with a separate enterprise tier. AGPL is a recognized OSI license, stricter and more copyleft than n8n’s fair-code license.
Self-host: Docker Compose, plus PostgreSQL and Redis. Setup is straightforward, comparable to n8n’s queue mode.
Where it wins: Simplicity. The Zapier-style linear builder is less intimidating than n8n’s node graph for non-technical users, and the privacy/GDPR self-host story is front and center.
Watch out for: It is thin and slow-moving. There are only about 80 connectors (versus n8n’s 400+), it is still pre-1.0, and releases are roughly annual - the latest, v0.15.0, landed in August 2025, though commits continued into early 2026. It added AI nodes (OpenAI, Anthropic, MCP/agents) in that release.
Best for: Non-technical or privacy-focused teams who want a simple, genuinely open-source (AGPL) Zapier replacement for a handful of common connections, not power users.
Code-First (own your automations as code)
4. Windmill - scripts into workflows, with auto-generated UIs
TL;DR: Code-first automation (AGPLv3) - write Python, TypeScript, Go, or Bash and get versioned jobs with auto-generated UIs and APIs. Best for engineering teams that want automations in Git, not a no-code canvas.
Windmill flips the model: automations start as code, not nodes. You write scripts in TypeScript, Python, Go, Bash, and more. Windmill turns each into a scheduled, versioned job with an auto-generated UI and API endpoint, then composes them into flows.
License: AGPLv3 for the community edition (Apache 2.0 for the API specs), with a source-available enterprise tier.
Self-host: Docker Compose brings up server, worker, and PostgreSQL in under a minute. Rule of thumb is about one worker per vCPU and 1-2GB of RAM. The engine is written in Rust and built for heavy, long-running scripts.
Where it wins: Every workflow is real code in Git, reviewable in pull requests and wired into CI/CD. It supports 10+ languages with full PyPI/npm imports, and it includes a Retool-style internal-app builder that n8n does not have.
Watch out for: Not for non-developers. It lacks n8n’s library of pre-built no-code integration nodes - you call APIs by importing SDKs in code. Smaller community (~17k stars) means fewer ready-made templates.
Best for: Engineering and DevOps teams that want to own automations as multi-language code with Git review, and also want auto-generated internal UIs and APIs. Extremely active, with multiple releases per week in 2026.
5. Trigger.dev - durable background jobs in TypeScript
TL;DR: Apache 2.0, TypeScript-only durable background jobs and AI agents as code, self-hosted with no run limits. For JS/TS developers, not no-code users.
Trigger.dev is an open-source platform for writing long-running background jobs, durable workflows, and AI agents entirely in TypeScript. You define “tasks” as functions in your own codebase and get durable retries, queues, concurrency control, scheduling, and human-in-the-loop approvals.
License: Apache 2.0 - fully permissive, no commercial-use restrictions.
Self-host: Docker Compose or an official Helm chart. It runs on PostgreSQL + Redis with a worker layer, so budget ~2-4GB of RAM. Self-hosting has no run-count limits.
Where it wins: Real developer ergonomics. Version-controlled TypeScript in your own repo, type safety, no execution timeouts, durable retries, checkpoints, idempotency keys, and first-class OpenTelemetry observability.
Watch out for: No visual editor and no integration library - everything is code. The self-hosted edition is deliberately feature-reduced versus their cloud (no warm starts, no auto-scaling, no checkpoints).
Best for: TypeScript and JavaScript developers who want durable background jobs and AI-agent workflows as code, self-hosted without run limits. ~15k stars, with frequent point releases since the v4 GA in mid-2025.
6. Temporal - crash-proof workflows for backend engineers
TL;DR: MIT-licensed durable execution for crash-proof, long-running backend workflows in Go/Java/TS/Python/.NET. Heavy to run and code-only - infrastructure, not a no-code tool.
Temporal is a durable execution platform. You write workflows and activities as ordinary code in a Go, Java, TypeScript, Python, or .NET SDK. The Temporal Service persists every step, so the workflow survives crashes, restarts, and outages and resumes exactly where it left off.
License: MIT for the self-hostable server and all SDKs. Temporal Cloud is a separate commercial product.
Self-host: Heavier than most here. The service is multi-component (frontend, history, matching, worker) and needs an external database (PostgreSQL, MySQL, or Cassandra), plus you run your own worker processes. Plan for ~3-6GB of RAM across the stack.
Where it wins: Reliability guarantees nothing else on this list offers - automatic state persistence, built-in retries and timeouts, saga/compensation patterns, and processes that can run for days, weeks, or months at high throughput.
Watch out for: This is backend infrastructure, not a no-code tool. No node editor, no pre-built connectors, steep learning curve. It is an “n8n alternative” only for the developer-orchestration use case.
Best for: Engineering teams building reliability-critical, long-running, or stateful backend workflows - payments, order fulfillment, sagas, AI-agent orchestration. ~21k stars, with server v1.31.0 shipped in 2026.
7. Huginn - the hackable, MIT-licensed web agent
TL;DR: An MIT-licensed, self-hosted IFTTT for watching the web (scraping, RSS, alerts). Mature and permissive, but no visual canvas and no tagged release since 2022.
Huginn is a Ruby on Rails system for building “agents that monitor and act on your behalf” - the project calls itself “a hackable version of IFTTT or Zapier on your own server.” It predates n8n by years.
License: MIT. Fully permissive.
Self-host: Official Docker image, plus a MySQL or PostgreSQL database and a background worker. The project recommends 2GB of RAM (it will limp along on 0.5GB plus swap). A fresh idle container sits around 300MB.
Where it wins: Mature, battle-tested web-scraping and monitoring agents (Website Agent, RSS, scheduler) wired into an event graph, all under a genuinely permissive license with no paid tier or cloud upsell.
Watch out for: It is showing its age. The last tagged release was August 2022 - the repo is kept current on dependencies by essentially a single maintainer, but there have been no headline features in years. Configuration is JSON option fields plus Liquid templating, not a visual canvas, and there are no AI nodes.
Best for: Developers and tinkerers who want an MIT-licensed, self-hosted agent that watches the web (scraping, RSS, alerts, scheduled checks) and do not mind hand-wiring JSON and Ruby. Popular and stable (~49k stars), but feature-stagnant.
Data and Infrastructure Orchestration (a different job than n8n)
These two are not no-code SaaS-automation tools. If your real need is data pipelines or scheduled batch jobs, they are the right tool - and people often land on them after outgrowing n8n for data work.
8. Kestra - declarative, YAML-defined orchestration
TL;DR: Apache 2.0 orchestration with declarative YAML flows and 1,200+ plugins, for data, AI, and infra pipelines. JVM-based (4GB+ RAM), engineering-focused - Airflow/Prefect territory, not no-code.
Kestra is an open-source orchestration platform where workflows are defined declaratively in YAML and executed event-driven, with a UI that round-trips to the YAML. Think Airflow or Prefect territory, with 1,200+ plugins.
License: Apache 2.0 for the core, with a separate proprietary enterprise edition.
Self-host: Docker or Docker Compose with PostgreSQL. It is JVM-based (Java), so the official minimum is at least 4GB of RAM and 2 vCPUs - meaningfully heavier than a small n8n instance.
Where it wins: Real orchestration at scale. Declarative YAML flows are Git-versioned infrastructure-as-code, with event-driven and cron triggers, backfills, retries, durable execution, and strong observability. Language-agnostic script tasks suit data and infra pipelines. The 1.0 line leans into “agentic orchestration” with AI-assisted authoring.
Watch out for: Not for non-technical users. You write YAML, the footprint is JVM-heavy, and consumer SaaS connectors are thinner than n8n’s.
Best for: Engineering and data teams that want Git-versioned, declarative, event-driven orchestration of data, AI, and infra pipelines. ~27k stars, shipping the 1.x line on a steady cadence in 2026.
9. Apache Airflow - the data-engineering standard
TL;DR: The Apache 2.0 standard for Python-defined data pipelines (DAGs). The heaviest option (4-8GB, ~9 services) and code-only - right for data engineering, wrong for app-to-app glue.
Apache Airflow is the long-standing standard for programmatically authoring, scheduling, and monitoring data pipelines defined as Python DAGs. It is a top-level Apache Software Foundation project.
License: Apache 2.0.
Self-host: The heaviest option here. The default Docker Compose stack spins up about nine services (scheduler, dag-processor, API server, worker, triggerer, Postgres, Redis, and more), and the docs ask for at least 4GB of RAM, ideally 8GB. You can trim it with the LocalExecutor, but baseline complexity is high.
Where it wins: Power for real data engineering - complex DAG dependencies, scheduling and backfills, dynamic pipeline generation in pure Python, horizontal scaling via Celery or Kubernetes executors, and a mature provider ecosystem (Snowflake, Spark, dbt, S3, and so on). Airflow 3 added multi-team isolation and human-in-the-loop approvals.
Watch out for: No-code it is not. Everything is Python, there is no visual builder, and it is the wrong tool for quick app-to-app glue or webhooks.
Best for: Data engineers building scheduled, Python-defined ETL/ELT and batch pipelines. ~46k stars, with Airflow 3.2 shipping in 2026.
Side-by-Side Comparison
| Tool | Style | License | Rough RAM | Best for |
|---|---|---|---|---|
| n8n (baseline) | Visual + code | Sustainable Use (fair-code) | 2-4GB+ | Widest integrations, AI workflows, visual + code |
| Activepieces | No-code visual | MIT | ~0.5-2GB | Open-source, AI/MCP-native Zapier alternative |
| Node-RED | Visual flow | Apache 2.0 | ~100MB | IoT, edge, lightweight, Raspberry Pi |
| Automatisch | No-code visual | AGPLv3 | ~1-2GB | Simple Zapier-style flows, privacy-first |
| Windmill | Code-first | AGPLv3 | ~1-2GB/worker | Multi-language scripts as Git-versioned workflows |
| Trigger.dev | Code (TypeScript) | Apache 2.0 | ~2-4GB | Durable TS background jobs, no run limits |
| Temporal | Code (multi-lang) | MIT | ~3-6GB | Crash-proof, long-running backend workflows |
| Huginn | Config / DIY | MIT | ~0.3-2GB | Web monitoring and scraping agents |
| Kestra | Declarative YAML | Apache 2.0 | 4GB+ | Data, AI, and infra pipeline orchestration |
| Apache Airflow | Python DAGs | Apache 2.0 | 4-8GB | Python ETL and batch data pipelines |
Which n8n Alternative Should You Choose?
Want a no-code visual tool like n8n, but truly open source? Start with Activepieces (MIT, AI-native). If you want something featherweight or IoT-focused, use Node-RED.
Want the simplest possible Zapier-style builder for a few connections? Automatisch - know it moves slowly and has a small connector library.
Want to own your automations as code in Git? Windmill if you want multiple languages plus internal UIs; Trigger.dev if you live in TypeScript.
Building reliability-critical, long-running backend workflows? Temporal for durable execution that survives crashes.
Orchestrating data or ETL pipelines? Kestra for declarative YAML, Apache Airflow for Python DAGs.
Want a lightweight web-monitoring agent on your own box? Huginn, with the caveat that it is feature-stagnant.
Or Just Self-Host n8n Itself
Here is the honest take: if your reason for leaving is the license or a need for code-first control, the alternatives above are genuinely better. But a lot of people search for “n8n alternatives” for a different reason. n8n Cloud’s execution-based pricing (from $24/mo for 2,500 executions) gets expensive. And standing up a production n8n with PostgreSQL, queue mode, and a reverse proxy is real work.
If that is you, you do not need a different tool. You need cheaper, simpler hosting for the tool you already like.
Deploying n8n on InstaPods is one click, fully configured, on a real server with HTTPS - $7/mo flat on the Build plan, with unlimited workflow executions and no per-task fees. That is the full n8n Community Edition (all 400+ integrations and code nodes), not a stripped-down version, and it undercuts n8n Cloud while skipping the DevOps. You also get SSH access to the server, so your workflows and Postgres data stay yours - export or move them off any time, no lock-in. See the full n8n pricing breakdown for the Cloud-versus-self-hosted math.
Deploy n8n on InstaPods - one click, $7/mo flat, full Community Edition on a real server with HTTPS.
FAQ
Is n8n actually open source? No. n8n is source-available under the Sustainable Use License (fair-code), not an OSI-approved open-source license. You can self-host and use it internally for free, but you cannot resell it as a hosted service. For a truly open-source license, Activepieces (MIT) or Node-RED (Apache 2.0) qualify.
What is the best open-source n8n alternative? Activepieces is the closest match: an MIT-licensed, no-code visual builder with around 280 AI- and MCP-native connectors. It runs on a 1.5-2GB VPS and gives you n8n’s feel without the fair-code license.
What is the lightest n8n alternative? Node-RED. It idles around 100MB of RAM and runs comfortably on a Raspberry Pi, far below n8n’s 2-4GB footprint. It is Apache 2.0 licensed and strongest for IoT and edge automation.
Is there a code-first alternative to n8n? Yes. Windmill turns Python, TypeScript, Go, and Bash scripts into workflows with Git versioning, and Trigger.dev does the same for TypeScript background jobs. Both let you own automations as version-controlled code instead of a visual canvas.
Are these n8n alternatives free? All nine are self-hostable for free; you pay only for the server. The licenses differ - Activepieces, Temporal, and Huginn are MIT; Node-RED, Kestra, Airflow, and Trigger.dev are Apache 2.0; Windmill and Automatisch are AGPLv3. n8n itself is free to self-host too, under the more restrictive Sustainable Use License.
Related reading:
- n8n Pricing 2026: Cloud vs Self-Hosted Costs - the real numbers, every plan compared
- n8n vs Make - if you are comparing n8n against the SaaS incumbents
- n8n vs Zapier - per-task pricing math, side by side
- Self-Hosted Apps Directory - one-click open-source apps you can run on your own server
- Open-Source Hosting - why self-hosting beats SaaS subscriptions for most teams