AI Agent node
An agent step inside a workflow, connected to a chat model, memory, and tools. Models from OpenAI, Anthropic, Google, and others can be mixed in one workflow, on your own credentials.
Compare
n8n
a workflow automation platform, self-hosted or in n8n's cloud, with an AI Agent node that uses integrations, workflows, and MCP servers as tools
Who they are and what they are good at
n8n is a workflow automation platform for technical teams. You build on a canvas of nodes, drop into JavaScript or Python where needed, and run it in n8n's cloud or on your own servers. Its AI features sit inside those workflows.
Tools
Any integration, a sub-workflow, an HTTP request, code, or an external MCP server through the MCP Client Tool. The MCP Server Trigger exposes n8n tools to outside agents.
Human review for tools
Since n8n 2.6, any tool an agent can call can require a person's approval first. The reviewer sees the tool and its parameters, then approves or denies.
Where it runs
n8n Cloud, or self-hosted. The Community Edition is free to self-host under the Sustainable Use License. Business is self-hosted only, and Enterprise runs either way, including fully on premises.
Pricing
Per workflow execution, regardless of step count. An agent turn counts as one execution, and sub-agent and tool calls add nothing. SSO starts on Business; audit logging and log streaming are Enterprise.
For teams with engineers, it is powerful and cheap at volume. A self-hosted n8n sits inside your network and can call internal APIs and databases directly.
Where n8n stops for messy back-office work
n8n reaches anything with an endpoint. A lot of back-office work runs through software that has none.
Desktop apps with no API
A Windows ERP, a terminal screen, or a desktop accounting package has no node. Self-hosting puts n8n near the system, not inside its screens.
Logins that belong to a person
A credential in n8n is an app connection. A supplier portal tied to one buyer's login, with MFA on their phone, is a different thing.
Who maintains it
A workflow of nodes, expressions, and code belongs to whoever can read them. That is often not the AP lead who owns the process.
Review with context
A tool approval shows the parameters the agent chose. It does not show the invoice image beside the screen the agent is about to change.
LaunchAI works through the screen of any app, on the user's machine or an enrolled agent PC, in sessions that are already signed in. Calculations run as rules with no model call, and every action leaves a screenshot. The person who does the work builds it by talking and showing.
Side by side
| n8n AI agents | LaunchAI | |
|---|---|---|
| Built for | Technical teams automating across APIs | Recurring back-office procedures, including desktop apps |
| Who builds | Developers and technical operators | The person who does the work |
| How you build | Nodes on a canvas, expressions, code | Talk and show; each step runs live as you build |
| Reaches software through | Integrations, HTTP, code, MCP | The screen of any app, plus an API where one exists |
| Where it runs | n8n Cloud, or your own servers | The user's machine or an enrolled agent PC |
| Desktop apps | No | Yes |
| Exact rules | IF and Switch nodes, code | Rules and Tables, no model call |
| Human review | Approval on agent tool calls, Wait node | A review step anywhere, source beside the record |
| Evidence | Execution data; audit logs on Enterprise | A screenshot of every action |
| Cost unit | Executions, or free to self-host | License |
| Speed | Fast on API calls | Human pace per step, the tradeoff for reach |
When to choose n8n
- You have engineers and want to own the stack, on your own servers.
- High-volume flows between systems with APIs, where one run counts as one execution however long it is.
- An agent that answers questions from internal databases and documents, with risky tools behind approvals.
- Data must stay inside your network, and every system in the job has an endpoint.
When to choose LaunchAI
- A distributor's vendor invoices arrive as PDFs and phone photos. Each one is keyed into a Windows ERP with no API and matched to the purchase order and the receipt.
- Price and quantity tolerances live in a table the AP lead edits, and the match runs with no model call.
- Anything outside tolerance stops for the AP lead, with the invoice image beside the ERP screen.
- The person who owns the process can change it next month without filing a ticket.
Better together
Keep n8n for intake and the API work. An n8n workflow collects invoices from email, files them in a shared folder, and logs each one in a table. LaunchAI picks up new files on its schedule, enters and matches them in the ERP, and writes the result back to the table for n8n to route onward.
FAQ
Doesn't self-hosting n8n solve the on-premises problem?
For systems with an API or a database you can reach, yes. For a desktop app with no API, n8n still has no way into the screen.
Can n8n's agents require approval before acting?
Yes. Since version 2.6, any tool connected to the AI Agent node can require a human review step. Approve runs the tool with the agent's input; deny cancels it.
Which models does n8n's agent use?
Your choice. n8n connects to OpenAI, Anthropic, Google, and other providers on your own credentials.
Is n8n open source?
Its source is public under the Sustainable Use License, which n8n calls fair-code. It allows internal business use, with limits on reselling n8n itself.
Which costs less?
For pure API work, n8n usually does. It bills per execution, and self-hosting the Community Edition has no execution fee. LaunchAI earns its license on the steps n8n cannot reach.
No prompts. No babysitting. No Integrations.