n8n automation that survives production.
Anyone can connect two nodes. The workflows that matter run thousands of times a week against real orders and real customers. We build n8n the way software is built: self-hosted, version-controlled, tested, monitored, with a human approval where the stakes are high.
What Gigabit does with n8n.
Gigabit designs, builds and runs automation on n8n: self-hosted installation, workflows that connect commerce, CRM, ERP and messaging systems, AI agent workflows, custom nodes and code where the built-in ones fall short, and migrations from Zapier and Make.
For a $28M multi-channel electronics seller, five automations on self-hosted n8n with custom Python and a small language model cut manual operations work from 32 hours a week to 4.5, and process 85% of returns without a person touching them.
The n8n work we take on.
The difference between a demo and a system is everything that happens when a step fails.
Self-hosted set-up
n8n in your own cloud with queue mode and workers, encrypted credentials, backups, staged upgrades and access control, so your data stays on infrastructure you own.
Workflow builds
Order, returns, inventory, reporting and vendor workflows across the systems you run, with retries, idempotency and a dead-letter path for what cannot be processed.
AI agent workflows
Agents built on n8n's AI Agent node with a deliberately small set of tools, grounded in your data, and tested against real past cases.
Custom nodes and code
Custom nodes for systems without one, and Python or JavaScript services alongside n8n for logic too complex to live in a canvas.
Migration from Zapier and Make
An inventory of what runs today, what it costs per task, and a rebuild of the workflows worth keeping on infrastructure with no per-task bill.
Monitoring and takeover
Execution logging, failure alerts and run-time dashboards, and rescue of brittle workflows someone else built.
n8n makes an agent easy to draw. Making it reliable is the work.
The canvas hides how many ways an agent step can go wrong.
Built into the platform
n8n has gone further than other workflow tools in treating AI agents as a first-class building block.
- An AI Agent node that calls tools you connect to it, with memory and a choice of model.
- MCP support in both directions: workflows can expose themselves as MCP tools, and an instance-level MCP server lets outside AI clients use them.
- A code node and a self-hosted option, so data and custom logic stay under your control.
What turns a workflow into a system
In the case study above, the exception router deliberately keeps a human approval in the loop. Knowing where to do that is most of the skill.
- An evaluation set of real past cases that every agent step must pass before and after each change.
- Approval steps on actions that move money or stock, and autonomy only where a mistake is cheap to undo.
- Structured outputs and validation between steps, so one odd model reply cannot corrupt a record.
- Complex logic moved into tested code, with n8n left to do what it is good at: orchestration.
Published prices, mapped to the job.
Every figure here is the same one on the pricing page. Nothing on this platform is quoted from a separate rate card.
A price marked “from” buys the base scope set out on the pricing page. Anything larger is quoted as a fixed price before work starts. Work with no natural end, such as ongoing development, is priced per engineer per month with no lock-in.
When n8n is the right call, and when it is not.
- You want automation on your own infrastructure, with data that does not pass through a third party.
- Volume is high enough that per-task pricing elsewhere has become a real cost.
- Someone technical, yours or ours, will own the instance.
- Nobody will look after a server. Use n8n's hosted plan, or stay on Zapier.
- The job is one simple app-to-app sync. A native integration is less to maintain.
- The core of the work is heavy computation or strict real-time processing. Write that as a service.
32 Hours Per Week Recovered: AI Workflow Automation for a Multi-Channel E-Commerce Operation
Five automations across Shopify Plus, Amazon, Walmart and wholesale, orchestrated in self-hosted n8n with custom Python for the complex logic and a human approval kept on order exceptions.
Questions n8n teams ask
How much does n8n automation cost with Gigabit?
One workflow, scoped, built and deployed with up to three tool integrations and error handling, is $8,000 fixed over three weeks. A production AI agent is a flat fee from $8,000. Several workflows or a migration are staffed as an embedded pod at $3,000–$7,500 per engineer per month. Running them afterwards is Managed AI Operations from $3,000 a month.
Should we self-host n8n or use n8n's cloud?
Self-host when data control, volume or custom nodes matter and someone will own the server. Use n8n's cloud when you want no infrastructure to manage and your volumes are modest. The workflows themselves are portable between the two, so it is not a permanent decision.
n8n or Zapier?
Zapier for a handful of simple connections owned by non-technical staff. n8n when workflows branch, loop or need custom code, when volume makes per-task pricing expensive, or when data should stay on your own infrastructure. Most companies we meet have outgrown Zapier for their core operations and still use it happily at the edges.
Can you fix n8n workflows someone else built?
Yes. We start by adding logging and failure alerts so the real error rate is visible, then fix the workflows that fail most or matter most: missing retries, duplicate processing, and agent steps with no validation.
Is an n8n workflow reliable enough for money and stock?
It can be, if it is built for it: idempotent steps, validation between stages, a path for records that cannot be processed, and a human approval on anything hard to reverse. In one production system we built, 85% of returns process with nobody involved and order exceptions still wait for a person on purpose.
Often scoped alongside.
Zapier & Make
An inventory of what is running, the fragile parts rebuilt, and a migration when the task bill stops making sense.
Zapier & MakeShopify
Builds, migrations, custom apps and Functions, plus support and returns agents grounded in live order data.
ShopifyHubSpot
Workflow automation, custom objects, integrations with billing and product data, and agents that span the CRM and beyond.
HubSpot
Name the workflow that eats the most hours.
Thirty minutes to scope it. One workflow is $8,000 fixed, live in three weeks.