Zapier vs n8n: breadth, control, and the volume math

What you are actually choosing between

Zapier is a proprietary cloud service with the largest connector catalog in the category — a catalog Zapier puts at 9,000+ apps and more than 66,000 triggers and actions as of 2026 — and it bills per task: every successful action step in a workflow consumes one. n8n is a source-available platform from Berlin that you can self-host free of charge for internal business use, and its cloud plans bill per execution: a single run of your entire workflow, regardless of how many steps.

Almost every other difference follows from those two facts. Who runs the infrastructure determines where your data lives and who patches vulnerabilities. What the meter counts determines what your invoice does when volume grows. So instead of a feature checklist, this comparison works through five constraints in order. The first one that binds usually decides.

All figures as of July 2026. n8n prices are EUR as displayed to an EU visitor; USD figures may differ.

Constraint 1: budget at volume

Start with the free tiers, because they set expectations badly in both directions. Zapier’s Free plan allows two-step Zaps only — one trigger, one action — with 100 tasks a month and 15-minute polling. You cannot prototype a ten-step workflow on it at all; multi-step Zaps arrive with the paid Professional plan. n8n Cloud, meanwhile, has no permanent free tier — only a trial. Its free path is the self-hosted Community Edition, which costs your time instead of your money.

Now the math that matters. Take a ten-step workflow — one trigger, nine action steps — running 10,000 times a month.

On Zapier, each successful action step consumes one task, while triggers, filters, paths, formatters, and delays are free. If all nine post-trigger steps call apps, that is nine tasks per run: 90,000 tasks a month. The entry Professional plan is $19.99 a month billed annually for 750 tasks — an effective rate near $0.027 per task. At that rate, 90,000 tasks would run roughly $2,400 a month. Treat that figure as an anchor, not a quote: per-task pricing improves at higher allocations (Zapier sells Professional tiers up to two million tasks a month), and the entry prices are the only ones verified here. The number also shrinks if your steps are the free kinds — a Zap where four of the nine steps are filters and formatters bills five tasks per run, not nine. But the structure holds: the invoice is billable steps times runs.

The overage mechanics deserve attention because they surprise people. Exceed your allocation and Zapier switches automatically to pay-per-task billing, capped at three times the base allocation, after which further runs are held. Held, not lost — but paused until you upgrade or a new cycle starts. And the pay-per-task rates change for billing cycles on or after July 15, 2026, so check your own billing settings rather than trusting any third-party table, including this one.

On n8n Cloud, the same workload is 10,000 executions — one per run, no matter how many steps or how much data. The Pro plan at €50 a month billed annually includes exactly 10,000 executions and 20 concurrent runs. Those are EUR prices as shown to an EU visitor in July 2026; annual billing saves 17% against monthly, and USD pricing may differ. Self-hosted, the marginal execution costs whatever your server does.

Resolution: for short workflows at modest volume, the entry plans are price-comparable. For multi-step workflows at real volume, the meters diverge by more than an order of magnitude at entry rates, and the gap grows linearly with step count. Count your billable steps before you choose.

Constraint 2: data residency

Zapier hosts customer data on AWS servers in the United States and does not offer EU-only data residency. It commits to GDPR and UK GDPR compliance and certifies to the EU-US Data Privacy Framework for transfers — an arrangement that satisfies many legal teams and fails others. Its assurance posture is SOC 2 Type II and SOC 3, audited annually, with reports available through its trust center; Zapier’s own compliance page does not list ISO 27001, PCI, or HIPAA.

n8n’s answer is structural rather than contractual: run it yourself, wherever your obligations require. The Sustainable Use License permits free self-hosting for internal business purposes, and the Business cloud tier additionally offers a self-hosted deployment option alongside SSO and Git-backed version control. n8n aligns its security program to SOC 2 with annual independent audits, sharing the SOC 2 report with enterprise customers through its trust center; ISO 27001 certification was not verifiable from official sources as of July 2026.

Resolution: if a contract or regulator says your data stays in the EU — or in your VPC, or on your own metal — Zapier is out of the running. This is the cleanest of the five constraints, because no plan tier changes the answer.

Constraint 3: technical bench

Zapier’s strongest property is that nobody has to run it. No servers, no upgrades, no patch windows, no capacity planning. Ops and RevOps people who do not write code assemble working automations from templates, and the entry plans cost less per month than an hour of engineering time. That is a real achievement, and for teams without an infrastructure bench it is the whole argument.

n8n asks more of you and gives more back. G2 reviewers consistently name the learning curve as the top drawback — error handling, sub-workflows, and code nodes take real ramp-up, especially for non-developers. Scaling a self-hosted instance means running queue mode: a main instance pushes execution IDs to Redis while worker processes pull and run them — genuine infrastructure work, with genuine capacity control in return.

And self-hosting means you own patch latency. Between December 2025 and January 2026, n8n disclosed four critical vulnerabilities in roughly two weeks, including CVE-2026-21858 — CVSS 10.0, unauthenticated remote takeover of exposed instances — with public exploits circulating. n8n shipped patches promptly; the burden of applying them urgently fell on self-hosters running exposed instances. That episode is not a reason to avoid n8n — disclosure and patching worked as intended — but it is the deal you sign when your automation layer becomes software you operate. If your team would not have patched within days, that team should not self-host an internet-exposed automation platform. n8n Cloud is the honest middle path: execution-based pricing with the patching handed back to the vendor.

Resolution: no engineers to spare, choose Zapier — or n8n Cloud if the pricing model matters more to you than the gentler on-ramp. A team that already operates services will find self-hosted n8n a modest addition with a large return in control.

Constraint 4: AI ambitions

Zapier’s agent story is Zapier Agents: describe the behavior in plain language and the agent gets triggers, knowledge sources, web browsing, and actions across the app catalog. Per the MIT AI Agent Index, the product reached general availability in December 2025. Three facts belong in any sober evaluation. First, Agents bill in “activities,” a meter separate from tasks — 400 a month on Free, 1,500 a month on Pro, with a 40-activity cap per run on paid plans. Second, per the MIT index, agents “do not have persistent memory between sessions”. Third, as of December 2025, Zapier described Agents as having “reached the final steps of SOC 2 readiness” — meaning the agent product was not yet inside the SOC 2 audit scope that covers the core platform. Around the agents sit two more meters: AI by Zapier steps bill by model tier from June 15, 2026 — 1x, 3x, or 5x tasks, or 1x with your own AI account — and Zapier MCP exposes the catalog to clients like Claude, ChatGPT, and Cursor at 2 tasks per tool call.

n8n’s approach is assembly rather than description. The AI Agent node is a LangChain-based cluster node: a root agent plus sub-nodes you choose for model, memory, and tools. Agents compose — one agent can invoke another agent as a tool. A native Evaluations feature offers AI-judged correctness scoring, string similarity, and custom metrics for tracking AI workflows in production, and MCP support runs in both directions: an MCP Server Trigger exposes n8n workflows to external clients, and an MCP Client Tool lets n8n agents call tools on external servers.

Resolution: if you want a chat-style assistant that acts across your SaaS stack with minimal setup, Zapier Agents get you there fastest — go in with eyes open about the activity meter, the memory limitation, and the compliance scope. If you are engineering an LLM pipeline where you pick the model, define the tools, and measure output quality, n8n’s node gives you substantially more surface to work with.

Constraint 5: governance

Run history first, because incident review and audit both depend on it. Zapier guarantees Zap history for a maximum of 60 days and displays up to 10,000 runs; only Enterprise accounts can customize retention. n8n Cloud is tighter at the bottom: Starter keeps at most 2,500 saved executions or 7 days, whichever prunes first; Pro keeps 25,000 for 30 days; Enterprise keeps 50,000 with no time limit. Neither entry tier survives a quarterly audit question, let alone an annual one. Self-hosted n8n flips the ownership: retention is whatever your pruning settings say, for better and worse.

Identity and access follow the same tiering. Zapier’s Team plan has included SAML SSO at no extra cost since January 28, 2026, with a new 25-user limit that pushes larger accounts to Enterprise. On n8n, SSO via SAML and LDAP, project-level permissions with custom roles, and external secrets managers arrive at the Business and Enterprise tiers; the free self-hosted edition does not include them, which surprises teams who assumed self-hosting meant everything.

Human approval gates exist natively in both. Zapier’s Human in the Loop action pauses a run for reviewers to approve, decline, or edit submitted data, and December 2025 added guest reviewers who approve through a secure link without workspace access. n8n pauses on Send-and-Wait operations across Slack, Gmail, Microsoft Teams, Telegram, Discord, and WhatsApp, with approval buttons, free text, or custom forms and configurable timeouts.

One governance item is legal rather than technical: n8n’s license. The Sustainable Use License allows free internal use and explicitly permits consulting — building workflows inside a client’s own instance — but forbids hosting one instance and charging clients to run automations on it, white-labeling, or offering n8n as a managed service. Agencies planning a hosting business on the free edition are planning a license violation. Embedding n8n in a product requires a commercial agreement whose pricing n8n does not publish.

Resolution: the governance ceilings of both products live in their top tiers. Decide whether Zapier’s shape — 60-day history, a 25-user SSO tier, Enterprise-gated retention control — or n8n’s shape — Business-tier-gated SSO, short entry-tier history, full control if you self-host — matches your obligations. And read the n8n license before building revenue on it.

Where Brahmalabs fits

Both products treat a workflow run as something that starts, executes, and finishes within the platform’s patience. Brahmalabs sits in adjacent territory: an AgentOps platform where every run is a durable Temporal workflow, so runs survive crashes and restarts, and a human-approval wait parks the run without burning compute. Approvals are a first-class node type — single approver or N-of-M consensus, SLA timeouts, escalation chains — rather than a notification pattern. Agent steps execute in sandboxed containers, and each run’s actual outbound network domains are captured and shown on the run page. Pricing is credit-metered by runs and sandbox-minutes, not per step or per seat, with EU data residency by default. Candor where it is due: Brahmalabs is a young product in early access, SOC 2 Type II is in progress rather than certified, and its integration catalog is hundreds of providers — far behind both Zapier and n8n on breadth, with MCP compatibility closing part of the gap. If your problem is app-to-app glue, stay with the two products above. If your problem has become running AI agents in production with governance attached, it is a different category of tool.

Where this leaves you

Advice by situation, as to a friend who will find out if it was wrong.

• You are an ops team without engineers, gluing long-tail SaaS together, and US data hosting is acceptable. Choose Zapier. The catalog is unmatched and the on-ramp is genuinely gentle. Count the billable steps in each Zap before committing, watch the usage bar, and re-run the volume math the day any workflow passes a few thousand runs a month.

• You run multi-step workflows at volume and have at least one technical person. Choose n8n — Cloud first. The execution meter is the right shape for your cost curve, and you defer the operational burden until residency or scale forces the move to self-hosting.

• Your data must stay in the EU, in your VPC, or on-premises. Self-hosted n8n is the only real option between these two. Budget honestly for patch latency: the December 2025 CVE cycle is what ownership looks like in practice.

• Your ambitions are mostly AI agents. Neither is a complete answer yet. Zapier Agents are the fastest route to an assistant that acts across apps — with a separate activity meter, no persistent session memory, and SOC 2 scope still catching up as of December 2025. n8n’s AI Agent node offers far more engineering control plus native evaluations, but you assemble and operate the result.

• You are an agency planning to resell automation. Zapier has no self-hosting to resell, and n8n’s license permits consulting inside client instances but not hosting a shared instance for paying clients. Read the license before you build the business plan.

Both products are good at the job they actually claim. The expensive mistake is buying one for the other’s job — paying task-metered prices for high-volume pipelines, or staffing an ops rotation for a self-hosted platform nobody needed. And if the job description eventually reads less like “connect apps” and more like “run agents under governance,” that is the point at which Brahmalabs is worth a look.