The same automations, on servers you actually control

Comparison

For teams that outgrew "it just works"

Run the same automations on hardware you own, with personal data stripped before any model sees it and a log that names the country each call reached. Zapier is the fastest way to connect two SaaS products and for a great many teams it is genuinely the right tool — it is also a hosted service, so your data crosses infrastructure you do not run to get from one app to the next. This page is for the point where that stops being acceptable, which is usually when the data becomes personal, regulated, or someone else’s.

Try the builder All comparisons

Where Zapier wins, and it is not close

Breadth and time-to-first-automation. Zapier connects to thousands of applications and gets a working automation running in minutes with no infrastructure at all. If you want to post new form submissions into a channel and never think about it again, running a container stack to do it would be a strange decision. Nothing on this page is an argument that Zapier is bad at what it does — it is an argument about where the work runs.

When teams start looking for something else

Almost always the same trigger: the automation starts touching personal data. A candidate CV, a patient reference, a customer complaint with a name and an address in it. At that moment the question changes from "does this work" to "who processed this, in which country, and can I evidence it". A hosted automation service can answer that with a policy. It cannot answer it with an egress log for infrastructure you do not run.

Moving across

What rebuilding actually involves

Bring the stack up

One command starts the workspace, the retrieval stack and the privacy guard. No licence key, no sales call, no card. This is the step that does not exist with a hosted tool, and it is the honest cost of the trade.

./selfhost.sh

Describe the automation you already have

Rather than dragging it out node by node, say what it does in a sentence and let the builder assemble a first version you correct. Most single-trigger automations come across in minutes because they were never complicated — they were just tedious to wire up.

When a form is submitted, summarise it and file it in the right folder

Connect the same accounts, then read the trail

Each person connects their own account by OAuth, and an integration acts as the person who authorised it. Afterwards, open the sovereignty view and read which country each call went to. That report is the thing you could not get before.

per-user OAuth · per-group permissions · egress by country

What you gain in the trade

Four things a hosted automation service structurally cannot give you

The automation runs on your hardware

Containers on your own machine or a data centre you chose, in a region you picked. The data does not transit a third party to get from one of your systems to another.

Personal data can be stripped before any model sees it

Install the detector and names, addresses, IBANs and national identifiers are tokenised on the way out and restored in the reply — running on your own CPU, not a detection API.

The AI in the automation is yours to choose

Any provider, or a model on your own network whose prompts never leave the building. Not a bundled assistant you cannot swap.

The integration list is not a ceiling

44 integrations ship built in, and then Bee Flow speaks the Model Context Protocol — so any MCP server becomes tools your automations can call, over stdio or Streamable HTTP. The public ones, the ones your vendors publish, and the one you write yourself for the line-of-business system nobody has ever built a connector for. Self-host and installing them is your decision rather than a request you file with a vendor and wait on.

Questions

What people ask when they are leaving a hosted tool

Can we import our Zaps?

No — there is no importer. In practice most single-trigger automations are quicker to describe in a sentence and correct than to convert, and the chat builder is designed for exactly that. Multi-step flows with a lot of branching take longer, and we would rather say so.

Do you have as many integrations?

Not as a catalogue — but the practical answer is open-ended rather than smaller. We ship 44 built-in integrations — Google Workspace, Microsoft 365, Nextcloud, ITSM, payroll, Dutch legal sources — and then the ceiling comes off, because Bee Flow speaks the Model Context Protocol. Any MCP server becomes a set of tools your assistants and automations can call, over stdio or Streamable HTTP: the public ones, the ones your own vendors publish, and the one you write in an afternoon for the line-of-business system nobody has ever built a connector for. When you self-host, installing those is your decision and nobody has to add your system to a catalogue first — the install surface is an enterprise feature, and once a server is installed, using it is an ordinary integration with the same permissions and audit trail as the rest. Where a large catalogue genuinely wins is the long tail of SaaS you want working in ten minutes without writing anything, so check the specific apps you depend on before you switch.

Is it harder to run?

Yes. That is the trade and it should be stated plainly: you are taking on a container stack, a database, backups and TLS, against a service where somebody else does all of that. The minimum is 2 CPU cores and 4 GB of RAM. If nobody in your organisation wants to own that, a hosted tool is the better answer.

What about the AI features?

They are not an add-on here. Assistants with memory, cited answers over your own documents, meeting notes and the privacy layer share the same permission model and audit trail as the automations, so a workflow can call an assistant and both appear in one trail.

Can we host it in the EU?

You choose the region entirely, because you choose the infrastructure. The platform also resolves every integration endpoint to a country at log time and flags transfers outside the EEA, so "is this data leaving Europe" becomes something you read rather than something you assume.

Rebuild one automation and compare the trail

Take the one that touches personal data. That is where the difference shows up.

Try the builder What it takes to run