n8n: the wiring around your agents
A self-hostable, fair-code workflow engine we use to trigger agent runs, plumb hundreds of integrations, and route human approvals — without shipping your data to anyone else's cloud.
- Self-hosted in your VPC
- 400+ native integrations
- Visual triggers & branching
- Human approval gates
A workflow engine, not a model
n8n is an open, node-based automation platform — think a programmable graph of triggers, integrations, and logic.
Each node does one thing: fire on a webhook or schedule, call an API, transform data, branch on a condition, or pause for a human. You wire them into a flow that runs on every event. Because it's fair-code licensed and self-hostable, the whole engine — credentials, payloads, and execution logs — stays inside your network.
We don't use n8n to make agents smart. We use it to make them connected and operable: it's the layer that decides when an agent runs, hands it the right tools and data, and catches what comes back. The reasoning lives in the model and orchestration layer; n8n is the nervous system around it.
What we use n8n for
The jobs n8n does well in an agent architecture — and the siblings it hands off to.
Triggers & ingestion
Webhooks, polling, schedules, and queue consumers that decide the moment an agent should wake up and act.
Tool plumbing
Hundreds of pre-built connectors to CRMs, ERPs, databases, and SaaS APIs — wired as the agent's action layer.
Approval gates
Pause a run for Slack or email sign-off before any high-stakes action executes, then resume on response.
Agent dispatch
Route requests to a LangChain or CrewAI agent, capture the result, and fan out follow-up steps.
In-perimeter execution
Self-hosted in your VPC or air-gapped so secrets and payloads never touch a third-party cloud.
Run observability
Every execution is logged with inputs, outputs, and errors — the raw material for audit trails and replay.
How n8n drives a run
A typical flow where n8n is the connective tissue around an autonomous agent.
Trigger
An event — a new ticket, a webhook, a schedule — fires the workflow and packages the context.
Dispatch
n8n calls the agent with the right tools and data, leaving the reasoning to the model layer.
Gate
High-risk actions pause for human approval via Slack or email before anything executes.
Act & log
The agent's actions run through n8n's connectors, and every step is recorded for replay and audit.
n8n or a durable engine?
n8n is built for event-driven glue: fast to author, visual to debug, and rich with connectors. That makes it ideal for the plumbing and the human-in-the-loop steps that wrap an agent.
It is not a durable execution engine. When a workflow has to survive for days, guarantee exactly-once semantics, or coordinate long-running state across many services, we move that backbone to Temporal — and keep n8n for the integration edges. We'll tell you which one your problem actually needs.
- n8n: triggers, integrations, approvals
- Temporal: durable, long-running state
- Airflow: scheduled batch pipelines
n8n vs. a custom integration layer
Why we often start with n8n instead of hand-rolling connectors.
| Hand-coded glue | n8n workflows | |
|---|---|---|
| Connectors | Build each one yourself | 400+ ready, plus generic HTTP |
| Iteration | Redeploy to change a step | Edit the graph, test instantly |
| Approvals | Custom UI and state plumbing | Wait-for-webhook nodes built in |
| Hosting | Your call | Self-host in your perimeter |
| Observability | Roll your own logging | Per-execution logs and replay |
Frequently asked questions
Is n8n the same as building an agent?
No. n8n is a workflow and integration engine — it triggers runs, plumbs tools, and routes approvals. The agent's reasoning still lives in a model (Anthropic, OpenAI, or open-source) and an orchestration layer. We use n8n as the connective tissue, not the brain.
Can we self-host n8n on our own infrastructure?
Yes — that's usually why we reach for it. n8n is fair-code licensed and runs in your VPC, on-prem, or air-gapped, so credentials and payloads never leave your perimeter. We deploy it behind your network controls with queue mode for scale.
When would you NOT use n8n?
When a workflow needs durable, long-running state with strict exactly-once guarantees and multi-day retries, we reach for Temporal instead. n8n shines for event-driven glue and fast iteration; it isn't a substitute for a durable execution engine.
Are we locked into n8n?
No. We treat n8n as a replaceable integration layer with logic and prompts kept portable. If you outgrow it, the agent core and action contracts move to Airflow, Temporal, or custom services without a rewrite of the intelligence.
Bring a workflow. We'll wire the agent.
One session to map your triggers, integrations, and approval points — and decide whether n8n, Temporal, or both belong in your stack.