Workflow automation for AI agents: where each platform stops being free

The short answer

The headline free-tier number on these pricing pages is close to useless, because no two platforms are counting the same thing. One counts every action against every record. One counts the whole run as a single unit no matter how much data flows through it. One counts seconds of compute. Comparing "100 tasks" to "1,000 credits" to "unlimited executions" is comparing three different currencies without an exchange rate.

There are only three families, and every platform in this category sits in one of them:

If your agent loops over a list — 50 rows, 200 tickets, 1,000 records — the per-run platforms are cheaper than the per-item platforms by roughly the length of the list. That is not a discount. It is a different unit, and it dwarfs every price difference on this page.

The same job, priced three ways

Make documents this better than anyone, in its own help centre, using its own example. A scenario watches a Google Form, then for each response creates a document, downloads it, and emails it. Ten responses arrive. Make counts: the trigger runs once for 1 operation, and each of the three downstream modules runs once per response for 10 operations each. Total: 31 operations for ten items.

The rule Make states is the important part: "bundles in earlier modules have a multiplying effect on the operations in the rest of the scenario." Your data volume multiplies your bill.

Now run the same shape at agent scale. Two hundred rows, two actions each — classify with an LLM, then write the result somewhere:

• Make: 1 (trigger) + 200 + 200 = 401 operations, by the rule above. For non-AI apps 1 operation equals 1 credit, and Make's free plan gives you 1,000 credits per month. That job runs twice before the free tier is gone. Modules with heavier processing or built-in AI connections are tagged as consuming 2 or 10 credits per operation, so the real figure can be several times higher.
• Zapier: a task is "any successful action performed in a third-party app after it's triggered". Triggers are free, and so are filters, paths and Formatter steps — Zapier only charges for actions sent out to other apps. Two hundred records with two outbound actions each is 400 tasks. The free plan includes 100 tasks a month. You are over budget a quarter of the way through the first night.
• n8n: "An execution is a single run of your entire workflow. It doesn't matter how many steps are in the workflow or how much data it processes." That is n8n's own definition, and it covers data volume explicitly. Two hundred rows through the same logic is one execution.

Same work. 401, 400, and 1. That is the whole buying decision, and it is decided before you look at a single price.

Activepieces uses the same per-run logic as n8n: "Running a flow costs 1 credit, however many steps are in it. AI steps cost more, 2 to 20 depending on the model, or 1 if you bring your own AI key." That last clause is worth reading twice — bringing your own API key drops an AI step from as much as 20 credits to 1.

Where each free tier actually ends

Two entries there deserve calling out.

n8n has no free cloud tier. Its pricing page offers free trials of Starter, Pro and Business — not a permanent free plan. Directories (including, until this article, our own listing) routinely record n8n as having one. It does not. n8n's genuinely free option is the self-hosted Community edition, which is a different product decision with a licence attached — see below.

Two vendors won't tell you the number. Pipedream's docs state that free workspaces have "a daily limit of free credits that cannot be exceeded" and link to a limits page that never states the figure; the pricing page renders entirely in JavaScript. Activepieces advertises credits that "refresh daily" without publishing the daily allowance. In both cases you discover where free ends by hitting it. That is worth knowing before you build something that matters on top of it.

What happens the moment you cross

This is the part that is genuinely undocumented across the category, and it splits cleanly into two behaviours: platforms that keep running and bill you, and platforms that stop.

They keep running:

• Zapier: "If you exceed your plan's task limit, you'll automatically switch to pay-per-task billing for the remainder of your usage period so your Zap workflows stay on."
• Activepieces, on the Plus and Team plans: "they keep running and you pay for the extra."
• n8n, on Business and Enterprise: "your workflows will continue running without interruption, but overage charges may apply." The Business overage rate is published — 4,000 EUR per additional bucket of 300,000 executions.
• Trigger.dev, on paid plans: "tasks keep running and compute past your included credits is billed at our compute usage rates."

They stop:

• Make: scenarios halt. Incoming webhooks queue up to your webhook storage limit and are "processed once your scenarios resume with added credits." Polling scenarios catch up from where they left off. You lose time, not data — provided the queue holds.
• Trigger.dev on the free plan: "you'll need to upgrade to keep running tasks once your $5 of free credits is used."

For an agent doing anything time-sensitive, that distinction matters more than the rate. A platform that silently converts you to metered billing cannot wake you up at 3am; a platform that stops can lose you a night's work. Pick deliberately.

The second cliff: the licence, not the invoice

For the self-hosted options, there is a boundary that never appears on a pricing page at all.

n8n's Community edition is source-available, not open source. The Sustainable Use License grants a free, worldwide licence with this limitation: "You may use or modify the software only for your own internal business purposes or for non-commercial or personal use. You may distribute the software or provide it to others only if you do so free of charge for non-commercial purposes." Files with .ee. in the name are excluded entirely and need an Enterprise licence.

Run n8n to automate your own company: fine, free, unlimited. Run n8n as the engine behind automation you sell to clients: not licensed. For an agency or a SaaS embedding workflows in its product — a large share of the people asking this question — that is the cliff, and it arrives at production, not at a usage threshold.

Activepieces does not have that restriction. Its licence is MIT Expat for everything outside packages/ee/. Commercial resale is permitted. GitHub's licence detector reports the repository as NOASSERTION because the file is a composite, which is why automated directories often get this wrong — read the file, not the badge.

Trigger.dev is Apache-2.0 across the repository, confirmed against the GitHub API today.

Both open projects are alive rather than nominally maintained: as of 15 August 2026, Activepieces sits at 23,793 stars with a commit pushed today, and Trigger.dev at 16,030 stars, also pushed today.

So which one

If your agent processes lists, self-host Activepieces. Per-run metering, MIT licence, no per-seat or per-task meter running while you iterate, and you can resell what you build. If you would rather not run infrastructure, its cloud free tier is the only permanent free cloud tier here that also meters per run.

If your work is internal and you want the largest ecosystem, n8n is the better tool and the worse licence. Self-hosted n8n is free and unlimited for your own company's automation, and its per-execution metering is the most generous model in the category. Just confirm you are inside the internal-use boundary before you build a business on it.

If your workflows are code and bursty, use Trigger.dev. $5 of compute a month free, Apache-2.0, and compute-second billing that doesn't care how many records you touch. The trade-off is that it bills for time, so a slow model costs more than a fast one for identical output.

Zapier and Make are the wrong shape for agent work at the free tier, and it isn't close. 100 tasks and 1,000 credits are demonstration budgets. Both are excellent at what they are actually for — broad app connectivity with thousands of maintained integrations — and if your automation touches one record at a time, Zapier's free tier is genuinely usable and its trigger-and-filter exemptions are the most generous in the category. But point either one at a 200-row batch and you are on a paid plan that same evening.

Pipedream sits between the families. Its per-30-second-of-compute model is explicitly not per-step — the docs say so directly, calling out "unlike some other platforms, Pipedream does not charge for usage based on the number of steps" — and public-registry source executions are included free. It is a strong model. We just cannot tell you where its free tier ends, because nobody publishes that.

What we could not verify

In the interest of not guessing: the current paid-tier prices for Make, Zapier and Activepieces are rendered client-side behind usage sliders, so there is no longer a single "Core is $X" figure to quote — the price is a function of the credit or task tier you select. We have deliberately left those numbers out rather than publish a figure we could not read from the source today. Every metering rule, free-tier boundary, overage behaviour and licence term above was checked against the vendor's own page or repository on 15 August 2026.