Self-Hosted n8n for Business: Setup, AI Agents and Workflows
n8n, self hosted, runs your AI agents and workflows inside your own cloud account, next to your data and internal systems. As an n8n automation agency, we set up queue mode, Postgres, SSO and backups, connect agents to self-hosted or API models, and hand over runbooks so your team owns every workflow.
Why Companies Self-Host n8n for AI Agents and Workflows
Companies run n8n self hosted when AI agents and workflows touch customer records, finance data or systems inside the firewall.
Workflow data stays in your network
Workflow definitions, execution history and files live in your Postgres database and storage, wherever you decide to host n8n (n8n pricing FAQ).
Credentials stay with you
n8n encrypts credentials with a key before saving them (n8n docs). That key, and the API keys for your CRM, ERP and databases, stay on infrastructure you control.
Direct reach into internal systems
Workflows and AI agents call internal databases, private APIs and file shares over private connections, under your existing firewall rules.
Model choice for every workflow
n8n connects to models from OpenAI, Anthropic and Google, and to models you host (n8n GitHub README). Sensitive steps run on self-hosted models; general steps call API models under your keys.
A licence your company holds
The Sustainable Use License covers use for your own internal business purposes (n8n LICENSE.md). Community is free to self-host; paid plans add a licence key for features such as SSO.
Access and retention on your terms
SSO through your identity provider, shared projects on paid plans, and execution data saved and pruned on your own retention rules.
n8n Hosting in Your Own Cloud Account
We self host n8n in containers on your AWS, Azure or Google Cloud account, or your Kubernetes cluster, following n8n's own hosting docs.
Containers on your infrastructure
n8n ships as a Docker image (n8n GitHub README). It runs behind your load balancer and TLS, with configuration in infrastructure code your team reviews.
Postgres as the database
n8n defaults to SQLite and also supports Postgres (n8n docs). A distributed setup needs Postgres, so we run it managed by your cloud provider or in your cluster.
Queue mode with Redis and workers
The main instance handles timers and webhook calls and passes executions to workers through Redis, the message broker (n8n docs). Workers scale out with AI agent volume, and n8n lists queue mode in the Community edition.
Shared storage for files
Queue mode needs binary data on shared storage such as S3, a paid self-hosted feature (n8n docs), so document-heavy AI workflows shape the edition choice.
Encryption key and secrets
All workers share one encryption key (n8n docs), which we hold in your secrets manager. Enterprise reads credentials from AWS Secrets Manager, Azure Key Vault, GCP Secrets Manager, HashiCorp Vault, Infisical or 1Password.
Backups and restore tests
We back up what n8n's guide lists: the database, the encryption key, external file storage and custom nodes, then prove each backup with a restore test.
Upgrades tested on staging
n8n's update guide advises updating at least once a month and reading release notes for breaking changes. Each upgrade runs on staging after a full backup.
Credentials, SSO and Access by n8n Edition
The edition sets which access controls your instance has. We match it to your SSO, secrets and logging needs, using n8n's docs and pricing page.
| Capability | Community (free, self-hosted) | Business (self-hosted only, under 100 employees) | Enterprise (self-hosted or n8n cloud) |
|---|---|---|---|
| Licence | Sustainable Use License | Paid licence key | Paid licence key |
| Queue mode with workers | Included | Included | Included |
| SSO (SAML, OIDC) and LDAP | Not included | Included | Included |
| Sharing workflows and credentials | Instance owner and creator | 6 shared projects | Unlimited shared projects |
| Environments and Git version control | Not included | Included | Included |
| External secret stores | Not included | Not included | Included |
| Log streaming to syslog, webhook or Sentry | Not included | Not included | Included |
| Multi-main high availability | Not included | Not included | Included |
| Support | Community forum | Forum support | Dedicated support with SLA |
Paid licence keys check in with n8n's licence server each day, and one key works on unlimited instances (n8n pricing FAQ), so we plan outbound access for staging and production together. Plan features change, so we recheck them before you buy.
For Claude, ChatGPT or another assistant calling your workflows, n8n's MCP Server Trigger exposes them as tools (n8n docs), and our AI integration services connect those assistants with per-user permissions and logging.
AI Agents in n8n on Self-Hosted Models
With n8n's AI Agent node, you connect a chat model and one or more tools, and the agent decides which tools to call (n8n docs).
Agent builds on your data
Our n8n developers build agents for intake, triage, research and record updates, each with a set tool list and accuracy checks on your own examples.
Self-hosted models through Ollama
The Ollama Chat Model node connects to an Ollama server at a base URL you set, so sensitive prompts and responses stay on your own GPUs.
vLLM through an OpenAI-compatible API
vLLM implements OpenAI's Chat API (vLLM docs), and n8n's OpenAI credential takes a custom base URL, so the OpenAI Chat Model node can call your vLLM server.
One LLM gateway for every workflow
Point every n8n agent at a self-hosted LLM gateway: one OpenAI-compatible endpoint with routing, fallback, per-team keys and budgets.
Human approval on tool calls
n8n can require a person to approve a specific tool call, with requests in Slack, Microsoft Teams, Gmail, Outlook and other channels (n8n docs). We put write actions behind that review.
MCP tools in both directions
The MCP Client Tool node gives agents tools from MCP servers, and the MCP Server Trigger exposes n8n workflows to external AI agents (n8n docs).
n8n Workflow Builds, Monitoring and Handover
Our n8n consulting covers the workflows too, and every build ends with your team owning it.
A map of workflows and data
Processes to automate, the systems and credentials each touches, the data classes involved, and which AI steps run on self-hosted models.
Builds tested on staging
Workflows are built on a staging instance with sample data. Business and Enterprise add environments and Git version control to move them between instances (n8n pricing page).
Accuracy checks for AI steps
Each AI step gets a test set from your own examples, so your team can measure extraction and classification accuracy before and after a model change.
Monitoring and alerts
Failure alerts go to the channel your team watches. Enterprise log streaming sends events to syslog, a webhook or Sentry (n8n docs).
A runbook for every workflow
What the workflow does, its credentials, its owner, and how to rerun a failed execution or roll back a change.
Named owners and admin training
Your admins practise users, credentials, upgrades and restores on your instance. For an n8n expert on call afterwards, you can hire an n8n developer from our team.
n8n vs Zapier for AI Workflows
Zapier is a hosted automation service with Free, Professional, Team and Enterprise plans billed on tasks, used when a Zap moves data or completes an action. It offers Agents, MCP, Copilot and Tables, and hosts customer data on AWS servers (Zapier pricing and data privacy pages). It suits:
n8n runs where your data and models already are.
Self-hosted n8n runs in your cloud account with your encryption key, database and choice of models, for AI agents that work with internal systems and sensitive data.
n8n vs Dify, Flowise and Langflow
Dify, Flowise and Langflow are builders for LLM apps and AI agents, while n8n is a workflow automation tool with AI Agent nodes. Here is how the four compare for a self-hosted build, from each project's GitHub repository and docs, checked November 2026.
| n8n | Dify | Flowise | Langflow | |
|---|---|---|---|---|
| What it builds | Workflow automation with AI Agent nodes, across 1500+ integrations | Agentic workflows and RAG pipelines in one collaborative workspace | AI agents and flows, built visually | AI-powered agents and workflows, served as an API or an MCP server |
| Licence | Sustainable Use License and n8n Enterprise License (fair-code) | Dify Open Source License, based on Apache 2.0 with added conditions | Apache 2.0 | MIT |
| Self-hosting | Docker image, with queue mode on Redis and workers for volume | Docker Compose, with at least 2 CPU cores and 4 GiB of RAM | Repository archived by its owner on August 13, 2026 and now read-only | pip or uv install, a Docker image or Langflow Desktop |
| Licence terms to check | Use for your own internal business purposes | Running a multi-tenant service from the source needs Dify's written authorization | The Apache 2.0 code stays yours to fork and maintain | Standard MIT terms |
| Best fit | Automations across your CRM, inbox and ERP with AI agents at the steps that need judgment | Chat, RAG and agent apps with their own interface, built by a product team | Existing Flowise installs planning a move to a maintained tool | Python teams that prototype agents and publish flows as APIs or MCP tools |
In the n8n vs Dify choice, the deciding question is where the AI work sits. When the job is an automation across business systems with an AI agent at one or two steps, the n8n AI Agent node covers it, next to the integrations your workflows already use. When the job is a chat or RAG app with its own interface, Dify fits; it also runs Dify Cloud, so a team can try it without managing a server (Dify GitHub README). Langflow vs n8n follows the same line: Langflow suits developers who publish flows as APIs or MCP tools, and either builder can run beside n8n in the same cloud account.
For Flowise vs n8n, the status of the project decides it. The Flowise team announced in July 2026 that it was winding Flowise down, said its npm packages and Docker images would be marked deprecated and archived the GitHub repository on August 13, 2026 (Flowise sunset notice and GitHub repository). Teams on Flowise keep the Apache 2.0 code they run today. We move those flows to n8n, Langflow or Dify when they are ready.
n8n vs LangGraph: When a Workflow Tool Is Enough
Anthropic's guide to building effective agents separates workflows, where LLMs and tools follow predefined code paths, from agents, where the LLM directs its own process and tool use. It recommends the simplest solution that does the job. n8n runs the first kind well, and LangGraph is built for the second.
| n8n | LangGraph | |
|---|---|---|
| What it is | Workflow automation with AI Agent nodes, built in a visual editor | A low-level orchestration framework for long-running, stateful agents, written in code |
| Licence | Sustainable Use License and n8n Enterprise License | MIT |
| Several agents | The AI Agent Tool node lets a root agent call other agents as tools | Agents and steps as a graph that you define and test in code |
| Human review | Approval before a specific tool call, requested in Slack, Microsoft Teams, Gmail or Outlook | People inspect and modify agent state at any point in a run |
| Long runs and recovery | Queue mode spreads executions across workers | Durable execution, so a run picks up where it stopped |
| Hosting | Self-hosted in your cloud account or on n8n's hosted plans | Your own servers or LangSmith Deployment (named LangGraph Platform until October 2025) |
| Who maintains it | Operations and automation teams, in the n8n editor | Engineers, in a code repository with tests |
A workflow tool is enough when the steps are known in advance: read an email, extract the fields, check the CRM, draft a reply and wait for approval. Most intake, triage and record-update jobs fit that shape. n8n lets your operations team see and change every step. LangGraph earns its place when an agent plans its own steps over a long run, resumes after an interruption partway through or keeps custom state that several agents share. The two also work together: n8n triggers the job and calls a LangGraph agent over an API for the open-ended part, built through our agentic AI development services.
For small and mid-sized businesses, our AI automation agency builds AI employees for the front desk, inbox and paperwork on self-hosted n8n first and adds code where a job calls for it.
What Shapes a Self-Hosted n8n Engagement
The scope follows your workflows, systems and data rules. Bring your process list, and we will map the n8n setup and AI agent builds that fit.
Book a Strategy Session →Frequently Asked Questions on Self-Hosted n8n
n8n describes itself as fair-code, distributed under the Sustainable Use License and the n8n Enterprise License (n8n GitHub README). The Sustainable Use License permits use for your own internal business purposes or for non-commercial or personal use, and source files with ".ee" in their name or folder need an n8n Enterprise License (n8n LICENSE.md). Your counsel confirms the fit.
SAML and OIDC single sign-on are on the self-hosted Business and Enterprise plans (n8n docs). The n8n Enterprise plan adds external secret stores, log streaming and multi-main high availability.
The Community edition is free to self-host. Business is paid, self-hosted only and aimed at companies with fewer than 100 employees; Enterprise is priced by n8n's sales team (n8n pricing page). Add cloud costs for containers, Postgres, Redis and any GPUs for self-hosted models.
Yes. The Ollama Chat Model node calls an Ollama server in your network, and the OpenAI credential's custom base URL lets the OpenAI Chat Model node call vLLM or a self-hosted LLM gateway. API models stay available for steps your data rules allow.
Your company does. You subscribe with n8n directly for a paid plan, and the instance, data and licence key stay in your name while we deploy and build on it. One licence key works on unlimited instances, per n8n's pricing FAQ.
Yes. As your n8n agency after handover, our developers take on new workflows, AI agent builds and upgrade testing inside your instance and under your access rules. Each new build arrives with its runbook and test set.
Related Self-Hosted AI Solutions and Services
Self-Hosted LLM Gateway for Model Routing and Cost Control
Learn more →Self-Hosted AI for Business: End-to-End Private AI Stack Deployment
Learn more →AI Integration Services for Claude, ChatGPT and Your Apps
Learn more →Ready to Run n8n in Your Own Cloud Account?
In a free strategy session, we'll map your workflows, systems and data rules, then outline the n8n setup and AI agent builds that fit.