Zapier costs $20 a month for 750 tasks. A task, in Zapier's billing model, is one step in one automation run. A workflow with three steps that runs 300 times a month burns 900 tasks and blows past that tier. Activepieces is a direct alternative, available at github.com/activepieces/activepieces with 23,494 stars, an MIT license, and a self-hosted path where the task counter does not exist. You run it on your own server, your automations run as often as they need to, and nobody charges you per step.
The practical pitch is simple: if you have automations living in Zapier or Make (formerly Integromat) and you are watching your monthly task count, Activepieces is where you move them. If you are running AI agents and need a layer that routes an agent's output to the right place without writing custom code, Activepieces fills that role too.
| Fact | Value |
|---|---|
| What it is | Open-source workflow automation platform |
| Built by | Activepieces, Inc. (open-source) |
| License | MIT (free to use and modify) |
| Platforms | Docker, cloud VPS, Activepieces Cloud |
| Requires | Docker for self-hosting |
| Install method | Single Docker Compose command |
| Verdict | Install if you need Zapier-style automation and want your data on your own server |
What Activepieces actually is
Activepieces is a visual workflow builder. You connect a trigger to one or more actions, and the platform runs the workflow every time the trigger fires.
A trigger is the event that starts the workflow. Examples: a new row appears in Google Sheets, a webhook receives a request (a webhook is a URL endpoint that receives data when an external event happens, like Stripe sending a notification when a payment completes), a form is submitted, or a schedule hits a specific time. An action is what the workflow does in response: send a Slack message, add a row to Airtable, create a HubSpot contact, call an external API, or run an AI step.
The connection between a trigger and an action is called a flow. Flows are built in a drag-and-drop canvas. Each step in the flow is a piece, which is the Activepieces term for a connector to a specific app or service. A connector is a pre-built integration that handles the authentication and API communication with a given service, so you never write code to connect to Google Sheets or Notion. You select the connector, authenticate with your account for that service, configure the step, and move on.
There are 200+ connectors available: Google Sheets, Gmail, Slack, Notion, Airtable, GitHub, HubSpot, Stripe, Shopify, HTTP requests, and many others. AI connectors for OpenAI and Anthropic Claude are included, which means AI steps slot in as ordinary steps in any flow rather than requiring a separate integration layer.
The self-hosted version runs on Docker. Docker is software that packages an application and all its dependencies into a container, a portable unit that runs identically on any server without manual configuration of the underlying system.
What you get after installing it
Once Activepieces is running, you work entirely in the browser dashboard.
The flow builder is the main screen. You pick a trigger piece from the left panel, configure it (for example: authenticate your Google Sheets account, select the spreadsheet and the sheet, choose "new row" as the event), then add action steps below it. Each action step can reference data from any earlier step using a point-and-click variable picker, so you can take the email address from a new Google Sheets row and pass it directly into a HubSpot "create contact" step without typing any code.
Branching logic ships out of the box. You can add an "if" condition step that routes the flow down different paths depending on whether a piece of data meets a condition. Loops are also available for processing lists, so you can take a Slack message with a list of items and have the flow process each item one at a time.
The AI steps are built in at the same level as every other step. An "Extract with AI" step can take unstructured text from an earlier step and pull out specific fields. An "Analyze Image with AI" step can take an image URL and return a description. A Claude or GPT step can take any prompt you define, with variables from earlier steps woven in, and return text that flows into the next action. These capabilities matter specifically for AI agent operators: when an agent produces a text result, a summary, or a classification, Activepieces can receive it via a webhook trigger and route it to a CRM record, a Slack channel, a database row, or a follow-up email without any additional code.
The dashboard also covers flow run history with a log of every execution, showing which steps succeeded, which failed, and what data passed through each step. Debugging a broken flow means clicking into a failed run and seeing the exact input and output of the step that failed.
The install experience
Self-hosting requires Docker and Docker Compose. Docker Compose is a tool that lets you define and start a multi-container application with a single command file, so Activepieces (which has a database, a background job runner, and a web server as separate components) starts as one unit rather than three separate manual steps.
The install is:
curl -fsSL https://raw.githubusercontent.com/activepieces/activepieces/main/docker-compose.yml -o docker-compose.yml
docker compose -p activepieces up -d
That pulls the official Compose file and starts the application. After a moment, the dashboard is available at http://localhost:8080. The first visit prompts you to create an admin account, and the flow builder is available immediately with no further configuration.
The only prerequisite is Docker. On a cloud VPS (a virtual private server, a rented Linux machine from providers like DigitalOcean, Hetzner, or Linode), you install Docker via one command, then run the two lines above. The whole process takes under ten minutes.
Activepieces also offers a hosted cloud version at activepieces.com if you want to skip the server step entirely. The cloud version has a free tier and per-task paid plans, which brings back the usage-based pricing. The self-hosted path is the one that removes it.
One operational note worth knowing upfront: self-hosting means you are responsible for keeping the server running and for updating Activepieces when new versions ship. Updates are a docker compose pull && docker compose up -d command. It is not complex, but it is a responsibility that does not exist with a hosted service.
Where it fits and what to compare it to
The honest comparison list is short and the choices are clear.
Zapier is the category leader: more connectors (7,000+), a polished user experience, and a hosted service with no server to manage. The tradeoff is cost. The $20 starter plan's 750 tasks disappear fast on any moderately active workflow, and the $49 plan for 2,000 tasks is still a ceiling. Activepieces self-hosted has no task ceiling, no per-step billing, and no restrictions on how often flows run.
Make (formerly Integromat) is cloud-only with a more visual interface than Zapier and a more generous free tier (1,000 operations per month). It is better for complex conditional branching but adds the same hosted-only limitation. There is no self-hosted option.
n8n is the other major open-source automation platform, also self-hostable via Docker, with around 100,000 GitHub stars. It has a larger connector library than Activepieces and is more powerful for technical users who want to write code inside flows. The tradeoff is a steeper learning curve. Activepieces is the more accessible choice for non-technical operators; n8n is the better choice if you want to drop JavaScript directly into your automation steps and need maximum flexibility.
Pipedream targets developers with a code-first approach and a free hosted tier. Not the right comparison for non-coders.
For a non-technical operator who wants Zapier-style automation without the per-task ceiling and without sending workflow data through a third-party cloud, Activepieces is the cleaner fit than n8n and the obvious replacement for Zapier costs.
Verdict
Install Activepieces if you are currently paying for Zapier or Make and the task count is a source of friction, or if you have AI agent workflows that need a routing layer between agent output and downstream systems.
The core experience is genuinely accessible. The trigger-and-action model is the same mental model as Zapier, which means there is almost no conceptual learning curve if you have used any hosted automation tool before. The 200+ connectors cover the apps that appear in most business workflows, and the AI steps integrate Claude and GPT at the same level as any other connector, which means adding an AI step to an existing flow takes the same two clicks as adding a Slack step.
The self-hosted MIT version has no usage limits and no licensing restrictions on commercial use. The main cost is a VPS at $5 to $10 a month, which for any team running more than a few active flows pays for itself quickly against a Zapier bill.
The one thing Activepieces does not yet have is Zapier's connector breadth. At 200+ connectors versus 7,000+, there are apps in Zapier that do not have a native Activepieces piece. The HTTP request step (a generic connector that sends a web request directly to any service's API, meaning the programming interface a web service exposes for outside communication) covers a large share of those gaps, but it does require knowing the specific URL and parameters the service expects rather than clicking through a pre-built connector. For the majority of workflows that live in Google Workspace, Slack, Notion, Airtable, HubSpot, or Stripe, that gap does not matter. For more obscure SaaS tools, check the connector catalog at activepieces.com before committing.
The project is actively maintained, the MIT license is clean, and the Docker install is reliable. It does the thing it promises.