Picking between Retool and n8n is usually framed as a tooling preference. It is really a question about the shape of the work in front of you, and the two pricing models pull in opposite directions hard enough that getting the shape wrong shows up on an invoice. Retool charges for people. n8n charges for volume. A quoting app used by twelve staff and a nightly reconciliation that touches 40,000 records are not the same problem, and the cheaper platform for one is the expensive platform for the other.
All prices below were observed on 18 October 2026 and are published in US dollars or euros. Australian-dollar figures are indicative conversions, marked as such.
The dividing line: a thing a human uses, or a thing that runs without one
Retool builds internal applications. Screens, tables, forms, buttons, role-based permissions, a login. Someone opens it, looks at data from three systems in one place, makes a decision, clicks approve. The value is in collapsing a process that currently lives across a CRM tab, a spreadsheet and an email thread into one interface that enforces the rules.
n8n builds unattended pipelines. A trigger fires, data moves between systems, transformations happen, something lands somewhere else. Nobody is watching. The value is in work that happens at 2am, or 400 times a day, without a person in the loop.
Most SME requests arrive described as "we need to automate X", and the first useful question is whether a human has to decide something in the middle. If yes, you need an interface, and that is Retool territory. If no, you need orchestration, and that is n8n. If the answer is "mostly no, but someone has to approve the exceptions", you need both, which is where most real builds end up.
What each one costs as published
n8n publishes a Business plan at €667 per month when billed annually, including 40,000 workflow executions per month with unlimited steps. At an indicative A$1.65 to the euro that is roughly A$1,100 per month. Enterprise pricing is not published.
The free path is real. n8n's documentation states the self-hosted Community edition can be used indefinitely and includes almost the complete feature set, with paid Business and Enterprise tiers adding SSO, environments, projects and governance features. Community is a genuine production option for a business that can run infrastructure. A third-party review estimates a basic VPS at roughly US$5 to US$10 per month, with a production PostgreSQL and Redis deployment at approximately US$20 to US$150 per month before any labour. That is an outside estimate, not an n8n list price, and the labour line is the one that matters: backups, upgrades, monitoring, security hardening and someone's phone ringing when a workflow stops at midnight.
Retool splits licensing between builders and internal users. Current pricing lists US$10 per builder per month and US$5 per internal user per month on Team, with Business at US$50 per builder and US$15 per internal user, on annual billing. At an indicative A$1.50 to the US dollar, a Team deployment with two builders and 25 internal users is roughly A$220 per month. The same shape on Business is roughly A$710 per month.
That split is the point. Retool is unusually cheap when a small number of people build apps that a larger number of people merely use. It gets expensive in the other direction — many builders, or a user count that keeps climbing, or external users who are a separate licensing question entirely.
The allowances most buyers discover late
Three numbers sit outside the headline licence and catch people.
Retool workflow runs. Retool's pricing lists 500 workflow runs per month on Free and 5,000 per month on Team and Business. Retool has scheduled and triggered workflows, and they are good, but 5,000 runs per month is about 166 a day. If your pipeline is a per-record trigger on a decent-volume queue, you will pass that quickly. This is the single most common reason a Retool-only architecture needs n8n underneath it.
Retool AI credits. The pricing page describes 250 AI credits on the free tier, with additional credits on paid tiers; credits are shared across the account, renew monthly and do not roll over. Shared and non-rolling is the part to model. An app that calls a model on every record view consumes a pooled resource that resets on a calendar, not on your workload.
n8n execution counting. n8n's published cloud pricing counts workflow executions, not nodes or steps — the Business plan includes 40,000 executions with unlimited steps. One customer record run through 20 nodes is generally one execution. One thousand records, each triggering the workflow individually, is one thousand executions. The same business outcome can differ by two orders of magnitude depending on whether you batch.
That makes trigger design a commercial decision, not just an engineering one. Before picking a plan you want an execution model: triggers per day, average records per trigger, retry behaviour, how often a schedule polls for changes that mostly are not there, and how many test runs development will burn. Estimating from "we have about eight automations" tells you nothing.
Where each platform stops
Retool stops at scale-out unattended work, at the workflow run allowance, and at the point where your users are customers rather than staff. Portal and external-user access is a separate cost line and should be priced separately, not assumed. Retool has also added a Self-hosted Business plan for teams needing granular permissions, audit logs and reusable modules. There is no reliable public list price for self-hosted Business or Enterprise, so treat it as quote-only rather than extrapolating from cloud tiers.
n8n stops at the interface. It has forms and basic input nodes, and they are fine for a trigger or a simple approval. They are not an application with a permission model, a searchable table of 60,000 rows, an audit trail of who changed what, and a layout that a non-technical staff member can work in all day. If the thing you are describing is a screen someone lives in, n8n is the wrong half of the stack.
Self-hosting n8n also does not solve compliance on its own. The documentation distinguishes Community, Business and Enterprise usage models clearly, but a self-hosted deployment still needs identity management, log retention, backups and a tested disaster recovery path. Australian hosting is a claim about where the application runs; it says nothing about backups, logs, support access or where model inference happens unless you ask each of those questions separately.
Why most builds use both
The common architecture for an Australian SME looks like this. n8n handles the ingestion and the movement — pulling from Xero, pushing to the CRM, watching a mailbox, calling a supplier API, retrying when it fails. Retool handles the exceptions and the oversight — the screen where an operations manager sees what failed, what is pending approval, and what needs a human decision, with the buttons that write back.
That split keeps each platform in its cheap zone. The high-volume work sits where volume is charged in blocks of 40,000. The human work sits where seats cost A$7.50 to A$22.50 a month. When we scope an internal application build on Retool for a client, the first design question is almost always which parts of the flow genuinely need a person and which parts are just moving data, because that line is where the architecture and the bill both get decided.
A rough decision table:
| What you are actually building | Start with |
|---|---|
| Screen replacing a spreadsheet process | Retool |
| Nightly sync between two systems | n8n |
| Approval queue with write-back to a CRM | Both |
| High-volume per-record processing | n8n, batched |
| Few builders, many occasional viewers | Retool |
| Lowest licence cost, you can run servers | Self-hosted n8n |
| Lowest operational burden | Vendor-hosted, either platform |
| Customer-facing product | Neither, price portal access separately first |
The honest part about effort
Neither tool makes the hard bits disappear. Delivery time goes into discovering a process nobody has written down, getting API credentials from a vendor who takes a fortnight to reply, reconciling source data that contradicts itself, deciding who can override an automated decision, and building the exception path for the 3% of records that do not fit. A narrowly scoped workflow on stable APIs moves quickly. A production workflow with authentication, retries, approvals, monitoring and audit logging is a different piece of work, and a quote priced by the number of screens or nodes is understating it.
Where to start
Take your top candidate process and write down two numbers: how many people touch it, and how many records flow through it per month. If people dominate, price Retool seats. If records dominate, model executions with your actual trigger design, including retries and polling. Those two numbers, honestly estimated, will tell you which platform is carrying the load and which one is the thin layer on top.
If you would rather not guess at the execution model, send us the process. We will map it to one architecture or the other, give you a monthly platform cost in Australian dollars based on your real volumes, a build scope with a two to four week delivery window, and a straight answer on whether the saving justifies the project at all. Tell us what the process is and we will come back with the numbers rather than a brochure.