Open Source Zapier Alternatives by Licence and Self-Hosting Cost

Why people leave Zapier

Per-task pricing. Zapier bills every action step as a task. Composio's June 2026 comparison works through the math: a five-step Zap that fires 150 times a month spends 750 tasks, the entire base Professional tier at $19.99 a month. Multi-step pipelines get expensive fast.

AI agents multiply the meter. The same Composio piece notes that routing an AI agent through Zapier MCP costs two tasks per tool call, so an agent that takes twenty steps to finish one job spends forty tasks. We've covered the wider shift away from per-seat and per-unit SaaS pricing in Per-Seat SaaS Is Dying. AI Agents Are Killing It.

Data residency and GDPR. Zapier has no general self-hosted edition (Sim, October 2026), so every Zap runs through Zapier's infrastructure. Self-hosting keeps workflow logic, credentials and execution history on servers you control, which is a much easier conversation with a compliance team.

The free tier won't hold you for long. Stackby's 2026 roundup puts Zapier's free plan at 100 tasks a month with single-step Zaps only.

Open source vs source-available: the licence table

"Open source" has a specific meaning: a licence approved by the Open Source Initiative (OSI) that lets anyone use, modify and redistribute the code, including commercially. "Source-available" means you can read and usually self-host the code, but the licence restricts some uses. As Sim's comparison puts it, self-hosting and open source are separate attributes.

Several tools also run an "open core" model: an OSI-licensed core plus cloud or enterprise code under a commercial licence. The licences below are as published by the sources named, checked October 2026. Open the LICENSE file in each repo before you commit, because projects do relicense.

Zapier advertises 8,000+ apps, and none of the open tools come close on catalogue size. That gap matters less than it looks, because the serious options all let you call any documented REST API through a generic HTTP request step. What you lose are the pre-built triggers and the OAuth plumbing, and that work lands on you. Before you pick, search each catalogue for the apps your Zaps depend on most, typically Slack, Google Sheets, HubSpot and Notion. Sim's comparison specifically advises checking the maturity of each connector you need, not just whether it exists.

Is n8n really open source?

No, not by the OSI definition. n8n uses its own Sustainable Use License, which opensourcealternatives.to notes is not OSI-approved; that site leaves n8n out of its OSI-only rankings for that reason. Robotalp labels it "fair-code (source-available)". You can read the code, modify it and self-host it, but the licence limits some commercial uses.

For a company automating its own operations, those limits are unlikely to come up. Composio notes that self-hosted n8n runs free with unlimited runs. Agencies and SaaS vendors planning to host n8n for clients, or to build it into something they sell, should read the licence text first and either talk to n8n about commercial terms or pick an MIT or Apache tool.

Tool profiles: strengths, wrong fit, catch

n8n

Strengths: The biggest integration catalogue of the self-hostable tools at 350+ (Stackby, 2026), JavaScript code inside visual flows (Lindy), native AI nodes (Goodspeed) and the largest community of the group.

Wrong for: Anyone reselling automation, given the licence, and teams where no one is comfortable running and maintaining a server. Composio flags that second point, since the self-hosted version is where most of the savings come from.

The catch: Paid tiers carry features the free self-hosted edition doesn't. Our sources don't list which ones, so if your security review requires SSO, granular role-based access or audit logs, confirm on n8n's pricing page which edition includes them before you build. If you'd rather not run it yourself, cloud plans start at $24 a month (Lindy, 2026).

Activepieces

Strengths: Several roundups, including Goodspeed's, call it the open tool closest to Zapier's simplicity, with a linear trigger-then-actions builder. Its MIT core is open source in the strict sense. Hosted pricing is the friendliest in the category: per its pricing page as summarised by Sim in October 2026, one credit equals one run regardless of step count. Run Composio's example through that: the five-step Zap firing 150 times a month costs 750 Zapier tasks but only 150 Activepieces credits, well inside the free 1,000.

Wrong for: Teams that need a long tail of niche connectors today. At 280+ integrations it trails n8n.

The catch: Open core. Sim notes that cloud and enterprise features use a commercial licence and tells teams to confirm which edition holds each capability they need. Check where SSO, audit logs and access controls sit before a company-wide rollout.

Automatisch

Strengths: AGPL-3.0, so open source in the strict sense. opensourcealternatives.to calls it the simplest no-code setup, with nothing to configure beyond Docker Compose, and pitches it at teams migrating a handful of Zaps.

Wrong for: Anything business-critical until you've verified the points below.

Is it still maintained? The most-cited review, from XDA in mid-2025, described it as pre-stable with a 15-minute polling interval for triggers, and the project's site footer has been reported as reading 2022-23. Our research doesn't show a current release state or commit history. Before you trust it, open the GitHub repo and check three things: the date of the latest commit, whether recent releases carry a stable version number, and how quickly issues get answered. If the last meaningful commit is months old, choose Activepieces, which covers the same no-code use case under an MIT core.

Node-RED

Strengths: Apache 2.0, light enough for a Raspberry Pi, with 4,000+ community nodes (opensourcealternatives.to). It is built for event-driven and IoT work.

Wrong for: Typical marketing-ops glue between HubSpot, Slack and Google Sheets. Orchestra's guide calls it less approachable for business teams than Zapier or n8n.

The catch: Most of that node count is community-built, so check who maintains any node your workflow depends on.

Huginn

Strengths: MIT-licensed and built from "agents" that watch websites, RSS feeds, webhooks and inboxes. n8n's own roundup recommends it for self-hosted web monitoring and scraping.

Wrong for: Non-developers. Robotalp describes the UI as configuration-heavy rather than a visual flowchart, and Activepieces' comparison calls it too technical, with an agent library that grows slowly.

The catch: A Ruby on Rails app to maintain, and new agents are written in Ruby.

Windmill

Strengths: Write scripts in Python, TypeScript, Go or SQL, and Windmill generates UIs and schedulers from your code (opensourcealternatives.to). For engineers, that beats wrestling with a visual builder.

Wrong for: Anyone who wants to click a flow together without code.

The catch: It assumes your team writes and reviews code, so the person who builds a flow is also the person who has to maintain it.

Apache Airflow and Apache NiFi

Both are data-engineering tools. n8n's roundup positions Airflow for orchestrating ETL and machine learning pipelines, and NiFi for moving and transforming data streams, with clustering to scale horizontally. Neither is designed around SaaS triggers like "new row in Google Sheets." Choose them only if your automation is ETL.

What doesn't belong on this list

Make, Gumloop, Lindy, Relay.app and IFTTT are hosted products whose code you can't run. Make is a reasonable budget alternative to Zapier, but it doesn't answer an open source question. Pipedream has stopped accepting Workflows signups and Sim reports a March 31, 2027 shutdown, so don't start new work there.

AI agents, MCP and local models

Composio's 2026 comparison says both n8n and Activepieces now ship AI agent and Model Context Protocol (MCP) features. Goodspeed highlights n8n's native AI nodes, and opensourcealternatives.to describes Activepieces as AI-native with MCP support. Robotalp notes Activepieces' Code piece includes AI assistance for writing JavaScript. On a self-hosted instance, agent tool calls don't run through a per-task meter the way they do on Zapier MCP.

Our sources don't document local-model connectors for any of these tools, so check the docs if you plan to run models on your own hardware. Budget for a much bigger machine too: a small VPS that runs the workflow engine comfortably won't run a model. Inference costs keep shifting, as we covered in Google's TurboQuant Cuts AI Inference Costs in Half, but GPU hosting is still a separate budget line.

Is n8n or Activepieces better for self-hosting?

For a technical team, n8n: more integrations, code steps, mature AI tooling and a larger community to troubleshoot with. For a non-technical team, or anyone who needs OSI licensing, Activepieces: easier to learn, an MIT core, and if self-hosting becomes a chore, its per-run cloud pricing is a cheap place to fall back to.

What self-hosting really costs

The minimum setup: a Linux VPS, Docker with Docker Compose, a persistent database such as Postgres, a domain with HTTPS through a reverse proxy, and a webhook URL reachable from the internet. opensourcealternatives.to puts initial setup at 10 to 30 minutes depending on the tool. The ongoing work costs more than the setup: applying upgrades without breaking flows, rotating credentials, watching disk space as execution logs grow, and testing that backups restore.

Who should not self-host: teams with nobody comfortable on a Linux command line, and workflows that need guaranteed uptime, such as payment handling or customer-facing order flows, unless someone owns the monitoring. A cheap VPS with no alerting can lose you a day of orders before anyone notices. If your automations feed a support bot, outages reach customers directly; see what that stack costs in What AI Customer Service Agents Cost Small Businesses.

Migrating off Zapier: the checklist

No tool imports Zaps directly. Every workflow gets rebuilt by hand. Sim's advice for non-technical teams is to rebuild one existing Zap first and confirm the app connections you need exist before moving the rest.

1. Inventory and prune. List your active Zaps and their task usage. Switch off the ones nobody would miss.
2. Rank by task cost. Move the expensive multi-step Zaps first, since that's where the savings are.
3. Map every trigger type. Zapier doesn't make it obvious whether a trigger polls (checks on a schedule) or uses a webhook (fires instantly), and your new tool may handle the same app differently. A trigger that polls every 15 minutes will break a workflow that assumed near-instant firing.
4. Create your own OAuth apps. This is the step people underestimate. On Zapier, "Connect Google" works because Zapier registered the OAuth app. On a self-hosted instance you often have to create your own OAuth client in Google Cloud Console, Slack's developer portal, HubSpot's developer account and so on, then set the redirect URL to your own domain. Google may also require app verification for some permission scopes.
5. Move credentials deliberately. Generate fresh API keys rather than copying old ones, and record where each one lives.
6. Rebuild filters, formatters and paths. Zapier's formatter steps map to code or utility steps. Test with awkward real data: empty fields, non-Latin names, duplicates.
7. Add error alerts. Send failure notifications to Slack or email from day one. Zapier emailed you when a Zap broke; your server won't unless you set it up.
8. Run in parallel. Point the new flow at a test destination, or run it alongside the Zap for a week, then switch over.
9. Turn on backups before you cancel Zapier, and test a restore.

Cost-driven platform switches are happening at every scale. For a vendor-level version of the same calculation, see Replit's move away from Anthropic.

FAQ

Is there an open source alternative to Zapier?

Yes. Activepieces (MIT core), Automatisch (AGPL-3.0), Node-RED (Apache 2.0), Huginn (MIT) and Windmill (AGPLv3) are open source under OSI-approved licences. n8n is the most popular self-hostable option, but its Sustainable Use License makes it source-available rather than open source. Zapier itself is proprietary.

What is the best open source automation tool?

For business app automation, Activepieces if you weigh licence and ease of use, n8n if you weigh capability and can accept a source-available licence. For code-first teams, Windmill. For IoT, Node-RED. For data pipelines, Airflow or NiFi. Base the choice on who will maintain it, not on the integration count.

What are the best alternatives to Zapier?

Self-hosted: n8n and Activepieces. Hosted and cheaper for multi-step work: Make (free 1,000 operations a month, Core from $10.59, per Stackby, 2026) or Activepieces Cloud (free 1,000 credits a month, Plus at $20, per Sim, October 2026).

Is Zapier still relevant?

Yes, for a specific buyer: someone who needs a niche app from its large catalogue, has nobody to run a server and keeps task volume low. It stops making sense once you pay for thousands of tasks on multi-step flows, route AI agents through its MCP meter, or face a data-residency requirement.

Can I self-host a Zapier alternative for free?

The software is free: the community editions of n8n and Activepieces run without per-task limits. The hosting isn't. Expect a small VPS bill, plus backups and someone's time for upgrades and fixes. See the cost table above.

Which open source Zapier alternative is easiest for non-developers?

Activepieces. Automatisch is also no-code and simple to deploy, but check its maintenance status first.

How do I migrate from Zapier to n8n or Activepieces?

Rebuild each Zap by hand, starting with the most expensive. Map polling and webhook triggers, set up your own OAuth apps for each service, and run old and new flows in parallel before you cancel. The checklist above covers each step.

Is Automatisch still maintained?

Our sources can't confirm a current stable release. The best-known review (XDA, mid-2025) called it pre-stable. Check the GitHub repo's latest commit date and release tags before you depend on it.