The 10 best low-code workflow automation tools in 2026

Most roundups of low-code workflow automation tools compare features and skip the question that decides the project: can you run it yourself, and on whose terms? Five of the ten tools below run only on the vendor's servers. Two ship source you can host but not resell. Three are open source. This list groups them by job and says what each license lets you do. We read the licenses in the repositories on 26 July 2026.

Short answers first, if that is all you came for.

• Widest app catalog, least effort: Zapier.
• Already paying for Microsoft 365: Power Automate.
• Self-hosted, widest integrations: n8n, if you can live with a source-available license.
• Self-hosted and open source: Activepieces.
• LLM agents on a canvas: Langflow or Dify.
• Nothing to keep running: build on Cloudflare Workflows, covered at the end.

Low-code workflow automation tools compared

These ten are the ones people shortlist. Star counts give a popularity signal for the projects with a public repository. The hosted platforms earn their place on market presence, which we cannot measure, so we have not ranked the two groups on one scale.

Proprietary, source-available, open source

Three tiers. The difference shows up after you commit.

Proprietary means the vendor runs it and you rent access. Zapier, Power Automate, Make, Workato and Pipedream all sit in this tier. Renting suits most teams. It also means your automations live somewhere you cannot inspect, priced on terms the vendor can change, and moving them later means rebuilding them.

Source-available means you can read and run the code but not do as you like with it. n8n ships under the Sustainable Use License, which the project calls fair-code: internal use is free, offering n8n as a hosted service to third parties is not, and the grant excludes every file with .ee. in its name. Dify uses a modified Apache 2.0 that forbids running a multi-tenant environment without written permission and forbids removing the Dify logo from the console. Most lists call both open source. The Open Source Initiative would approve neither license.

Open source means an OSI-approved license, unmodified. Activepieces (MIT), Node-RED (Apache 2.0) and Langflow (MIT) qualify. Activepieces carves out a packages/ee/ directory under commercial terms, the open-core pattern: the repository is open, but the feature you saw in the demo may sit outside the code you cloned.

Ignore all this if it does not affect you. If you plan to host automations for your own customers, read that column first. It rules out more of this list than the feature comparison will.

Hosted platforms

Zapier

The category default, and still the right first answer for most teams. It has the largest integration catalog in the industry by a wide margin, the editor needs no explanation, and a useful automation takes minutes. The limits arrive later: multi-step logic gets awkward, per-task pricing turns high-volume work expensive, and nothing runs on your own infrastructure.

Microsoft Power Automate

If your company runs on Microsoft 365, you already pay for this and IT has already approved it, which beats most technical arguments. It is strongest inside the Microsoft estate: SharePoint, Outlook, Teams, Dataverse, and desktop RPA flows for legacy apps with no API. Outside that estate it weakens, and the licensing split between seeded and premium connectors is the usual Microsoft puzzle.

Make

The most visual of the hosted tools. Make draws a scenario as a branching map rather than a linear step list, which makes complex routing and iteration easier to follow than the equivalent Zap. Cheaper than Zapier at volume, with a steeper first hour. See our Make comparison for the details.

Workato

The enterprise answer: governance, environments, audit trails, SSO, and a sales process to match. IT teams integrating systems of record choose it; individuals wiring up two SaaS apps do not. Priced accordingly, and not a tool you adopt on a Tuesday afternoon.

Pipedream

The developer-shaped member of this tier. Steps are Node.js, Python, Go or Bash over a large connector library, so it suits people who would rather write ten lines than fill in a form. It handles the plumbing; it will not turn a non-developer into an automation builder.

Self-hostable builders

n8n

The most popular self-hostable option by a distance, at 198k GitHub stars. Several hundred integrations, a real visual editor, JavaScript and Python code steps, and AI nodes built on LangChain. It runs as a long-lived Node.js process, so self-hosting means a VPS, Docker host or Kubernetes cluster that stays up, a Postgres instance beside it, and somebody to watch both.

Pick it when integration breadth matters and the license does not. Skip it if you intend to host workflows for customers. Our n8n comparison covers the license and runtime differences in full.

Activepieces

The MIT answer to the same question. Pieces are typed TypeScript packages, so writing a connector resembles publishing an npm module more than filling in a form, and the project ships MCP servers for agent use. The catalog is smaller than n8n's and the enterprise directory sits outside the MIT grant, but the core you clone is MIT.

Node-RED

Older than most of this list, Apache 2.0, and still the right tool for event-driven and device-driven flows. It came out of IBM's IoT work and that is where it stays strongest: MQTT, serial, hardware, home automation, edge gateways. Ask it to serve as a SaaS integration platform and it will disappoint you.

AI-first canvases

The fastest-growing branch of the category. Langflow and Dify hold more GitHub stars between them than every other project on this list combined, and they optimize for prompt chains, retrieval and tool-calling agents rather than for SaaS connectors.

Langflow

MIT, Python, 152k stars, and a drag-and-drop canvas for LLM chains and agents that exports to an API. The more permissive license of the two, and the easier one to embed elsewhere. Compare it with Dafthunk if you want the same canvas without a Python service to host.

Dify

The more complete product: workflows, RAG, datasets, observability and a publishable app layer, with 150k stars behind it. Its license repays a second reading, since running it multi-tenant is what the added conditions forbid. Our Dify comparison covers the rest.

Cloudflare Workflows, the layer underneath

Every tool above answers one question: which builder should I use? A second question underneath decides far more about what the automation costs to operate. What engine runs the steps, and what has to stay switched on for it to work?

The hosted platforms hide that engine and rent it to you. Every self-hostable tool on this list answers with a process running around the clock, a database beside it, and a person on call. That is the price of free software in this category, and it is why "we will just self-host n8n" turns into a small infrastructure project.

Serverless is the third answer, and serverless durable execution is rarer than the word suggests. AWS Step Functions and Azure Logic Apps qualify, but both orchestrate calls out to compute you provision and pay for somewhere else. We think Cloudflare Workflows is the strongest foundation available today. The engine and your own code run in the same runtime. Storage, queues and model inference are bindings on that platform rather than services you wire together across a network. A workflow can sleep for weeks and resume where it stopped, and an idle deployment costs nothing. Nothing stays alive between runs: no server, no container, no database.

That is the bet Dafthunk takes. It puts a visual canvas over Cloudflare Workflows: nodes execute as durable steps, state lands in D1 and R2, inference goes through Workers AI or your own OpenAI, Anthropic and Gemini keys, and you can expose any node as a tool to an agent. It ships MIT throughout, with no enterprise directory, and it deploys to your own Cloudflare account.

Dafthunk is young and smaller than anything ranked above, so we left it out of the list rather than rank ourselves beside tools with a decade of history. It inherits a constraint from its foundation: it ties you to Cloudflare, which rules it out for air-gapped deployments. If the trade suits you, browse the node reference, the workflow templates, or core concepts.

How to choose a workflow automation tool

Four questions settle it in most cases.

1. Who edits the workflow? If the answer includes people who do not write code, you need a visual builder, and Pipedream and the code-first engines drop out whatever else they offer.
2. Where does the data live? Regulated data or a hard residency requirement rules out the hosted tier, whatever the feature comparison says.
3. What will you do with it? Internal use makes the terms column almost irrelevant. Hosting automations for your own customers makes it decisive, and rules out n8n and Dify.
4. Who runs it at 3am? Every self-hosted option here needs a process, a database and someone on call. Price that before you price the licenses.

Questions people ask

What is the difference between low-code and no-code automation tools?

No-code means you build the whole workflow through a UI, as with Zapier or Make. Low-code means the UI covers the common path and you drop into code for the rest, as with n8n, Pipedream or Dafthunk. Almost every tool marketed as no-code keeps a code escape hatch, because real automations need one sooner or later.

Which workflow automation software is free?

Most hosted platforms offer a free tier capped by task volume. The self-hostable tools cost nothing to license for internal use, but you pay in hosting and operations. The cheapest total cost is usually the tool your team can already run.

Which of these are open source?

Activepieces, Node-RED and Langflow ship OSI-approved licenses. n8n and Dify are source-available, and most lists mislabel them. The other five are proprietary.

Which are easiest to self-host?

Node-RED and Activepieces are the lightest to stand up. n8n and Dify need more moving parts. Building on Cloudflare Workflows removes the question, because no host exists to manage.

Notes on this list

We read the licenses in each project's LICENSE file and pulled star counts from the GitHub API on 26 July 2026. We quote no prices and no connector counts, because both go stale faster than we update this page. Dafthunk is our product, so we describe it at the end rather than rank it. If a claim here is wrong or has aged badly, open an issue and we will correct it.

Related reading: Dafthunk compared with n8n, Zapier and Make, and building effective agents on Dafthunk for the agent patterns behind the canvas above.