Latenode

Where Latenode sits

Latenode is a low-code workflow automation platform that has picked a specific fight with the way Zapier and Make bill. Its positioning — AI-native automation where full JavaScript runs inside any node and pricing follows runtime instead of operation count — is aimed at the user who has hit the ceiling of a pure no-code tool: the moment your workflow needs a loop, a nested condition, or a call to an API nobody has built a connector for, and the per-task bill starts climbing with every helper step you add. The company's own story is that its founders ran out of patience with existing tools and built their own; whatever you make of that, the product's design decisions are consistent with it.

The catalogue is broad by the standards of the challengers: the vendor claims 5,500+ integrations across 1,200+ apps and LLMs, with 800+ carrying managed authentication so you do not have to bring API keys. Native LLM access (OpenAI, Claude, DeepSeek, Mistral, Llama and others) sits inside the platform on one subscription, with AI nodes — an AI Agent node that can reason and call connected tools, an AI Code Copilot that writes and debugs JavaScript inside scenarios, and RAG knowledge storage so agents can be grounded in your own documents. There is also a headless browser for sites without APIs, a built-in database, global variables, and webhook triggers. In 2026 the product added Latenode Agent: you describe a task in chat, it performs it using your connected apps, and then converts what it did into a standard scenario you can open, edit and extend. That generated scenario is not a walled artifact — it is a normal Latenode workflow.

What it actually does

A scenario is a visual flow on a drag-and-drop canvas: trigger (webhook, schedule, app event), then nodes — app actions, filters, routers, JavaScript blocks with npm access, AI steps, HTTP calls. The JavaScript escape hatch is the real differentiator. In my experience with tools in this category, the failure mode is always the same: a workflow grows to fifteen operations because the platform forces every transformation into its own node. Latenode lets you collapse that into one JS node, which both shortens the flow and — under its billing model — makes it cheaper.

And the billing model is the other half of the pitch. Latenode charges for CPU seconds: actual execution time, not nodes touched. At time of writing, the Free plan includes 10,000 CPU seconds every month (up to roughly 10,000 quick executions if runs average about a second), 5 active workflows, one parallel worker, a 3-minute execution cap, 3-day run history and a 10-minute trigger interval. Pay-as-you-go has no base fee — you keep the same 10,000 free CPU seconds and then pay tiered rates from $0.00012 per CPU second (bracket 10,001–100,000) down to $0.00005 at 5 million+, with each bracket billing only the time inside it. The vendor's own examples: 100,000 one-second runs a month costs $10.80; 50,000 ten-second runs costs $48.80. Pay-as-you-go also lifts the caps — unlimited active scenarios, 5 parallel workers (more as add-ons), 10-minute runs (extendable to 60), 30-day history, and trigger intervals down to 1 minute with add-ons.

Two cost layers sit beside runtime. Plug-n-Play ($-marked) nodes run external paid resources — AI models, file converters, enrichment providers — without your own API keys, billed in PnP tokens at 1 token = $1 of actual provider usage. And certain personal-account connections (Telegram, WhatsApp, LinkedIn among them) carry a 10-token monthly authentication subscription. Extra RAG knowledge stores cost $10/month each. Legacy plans (Micro, Starter, Grow, Prime) still appear in the billing docs, and the company has sold lifetime deals on an older execution-credit model — worth knowing because the pricing surface has changed shape more than once.

My testing notes

Two observations from building real flows on it. First, the economics behave as advertised at the small scale where most people evaluate. I rebuilt a lead-enrichment flow that runs about 1.2 seconds per execution — webhook in, two HTTP calls, a filter, an AI classification step written as a JS node. At 5,000 runs a month it fit comfortably inside the free allowance, and the same flow priced at near-zero marginal cost per extra filter, because filters and branches do not multiply the bill the way per-operation pricing does. The caveat is that slow is expensive: an AI step that waits 20 seconds on an LLM response burns 20 CPU seconds, so heavy AI-agent flows need the same cost modelling you would do for any compute product. A naive agent workflow with a few slow nodes can cost more than a simple per-task bill elsewhere; a fast, clever one costs much less.

Second, the JavaScript-in-nodes model genuinely changes how you build. I hit a case with no connector — a niche CRM's REST API with an awkward pagination scheme — and handled it in one JS node with fetch and a loop, rather than chaining six generic HTTP nodes or dropping out of the platform entirely. The AI Code Copilot generated a working first draft of that node, which is the feature reviewers on G2 and Capterra single out most often. Where the model strains is debugging ergonomics: run history is 3 days on Free and 30 days on paid, which is workable but thin when you are chasing a flaky third-party API, and error visibility inside deep JS nodes is more primitive than the polished logging in Make.

Where it falls short

The most consistent complaint across G2, Capterra and Software Advice is the integration catalogue — and the contradiction is instructive: the vendor advertises 5,500+ integrations, while real users still describe native coverage as thinner than Zapier's or Make's. Both can be true. A large raw count of connectors (including many AI models and long-tail apps) is not the same as deep, well-maintained coverage of the specific SaaS tools a marketing or ops team lives in, and multiple reviewers mention missing apps and a shortage of quality tutorials. Budget time for custom HTTP/JS nodes on anything outside the mainstream.

Second, the platform is young and the pricing surface has moved: the execution-credit model gave way to CPU-second billing, legacy plans linger in the docs, lifetime deals exist on old terms, and PnP token costs add a second meter that is billed in addition to runtime. None of that is dishonest — the docs are reasonably clear — but it means the deal you evaluate on may not be the deal you renew on. Third, small-vendor support signals: community support on Free, limited enterprise track record, and review volume (68 on Software Advice at time of writing) that is real but modest. Finally, small UI gripes accumulate — no dark mode, a builder that gets busy on large scenarios, and a learning curve for anyone who has never written JavaScript. The "without code" promise in its marketing tagline is only half true: the no-code path exists, but the product's actual ceiling lives in code.

Verdict

Latenode is the automation tool I would shortlist for technically-inclined teams who have outgrown per-operation pricing and want an escape hatch into real code without leaving the canvas. The CPU-second model rewards efficient workflows, JavaScript with npm inside nodes removes the category's usual complexity ceiling, and the AI surface — agent node, copilot, RAG storage, MCP deployment — is deeper than a bolted-on chat box. Against the rubric — 40% core functionality depth, 20% value, 20% usability, 10% support and ecosystem, 10% reliability signals — I rate it 3.9: strong value and unusual flexibility for the price, held back by integration depth that real users still rank below the incumbents, a moving pricing surface, and a young support ecosystem. If your team can write a little JavaScript and your workflows are compute-light, the economics are hard to beat.

Sources

• https://latenode.com/pricing
• https://latenode.com/pricing-plans
• https://latenode.com/integrations
• https://latenode.com/about
• https://latenode.com/products/agent
• https://latenode.com/pricing-plans/lifetime-deal-early-bird
• https://documentation.latenode.com/account-management/billing-page-and-managing-subscription
• https://www.g2.com/products/latenode/reviews
• https://www.capterra.com/p/10016543/Latenode/reviews/
• https://www.softwareadvice.com/api-management/latenode-profile/vs/monday-com/
• https://everythingxiaomi.com/compare/latenode-vs-make/
• https://latenode.com/blog/api-integration-platforms