n8n AI Agent

The n8n AI Agent is the best option on this list for technical teams that want AI agents wired into real business processes, with the choice to run everything on their own servers for free. The catch is skill: n8n gives you a canvas, a set of nodes and an agent builder, not a finished assistant, so someone has to design the triggers, tools and guardrails. Cloud plans start at €20 a month billed annually for 2,500 workflow executions, and the self-hosted Community Edition costs nothing.

It fits operations engineers, automation agencies and developers who already think in workflows, and it has momentum behind it: SAP invested at a $5.2 billion valuation in May 2026 and plans to offer n8n inside its Joule Studio agent builder. It fits a non-technical team that wants an assistant working by Friday much less well. This review is based on n8n’s published release notes, pricing terms and legal pages as summarized by independent pricing trackers in September and October 2026. We have not tested it hands-on.

n8n AI Agent by the Numbers

The company behind the n8n AI Agent was founded in Berlin in 2019 and more than doubled its valuation in under a year, to $5.2 billion, with SAP joining its cap table. The figures below cover its funding, growth, community and its most serious security flaw to date, each with its source and date.

The SAP deal is the most useful signal for enterprise buyers. A multi-year commercial agreement to run n8n inside Joule Studio means the agent canvas you build on today is also being adopted by one of the largest business software vendors, which reduces the risk of picking a niche tool. The growth figures and GitHub numbers point the same way: a large community means more templates, more answered forum questions and more people you can hire who already know the product.

The Ni8mare flaw is the other side of self-hosting. n8n fixed it, but every self-hosted instance only became safe once its owner upgraded. If you run the Community Edition, someone has to own patching, and that job belongs in the cost estimate.

How n8n AI Agent Works

n8n is a workflow automation platform. The AI Agent is a node, and increasingly a dedicated agent builder, that lets a language model decide which tools to call inside a workflow you design. Everything around it can stay deterministic: fixed steps, approvals and error handling.

Building an Agent, Step by Step

1. Choose where it runs. Sign up for n8n Cloud, which has a trial with no card required, or install the Community Edition with Docker on your own server.
2. Add a trigger. A chat message, webhook, schedule, new email, Slack mention or an event in another app. Since n8n 3.0, Slack agents can be set to reply only when mentioned.
3. Configure the agent. Pick a model provider, write instructions and attach tools: database queries, CRM lookups, HTTP calls to any API, other workflows, web search, MCP servers and Browser Use, which became available to all users on October 6, 2026. Memory sub-nodes keep context between turns.
4. Set limits. Monthly budget and per-session spend caps for agents arrived in October 2026, along with switches for individual capabilities and a setting that forces a tool call on the first step.
5. Test and publish. The agent builder’s Test agent button, an agent setup checklist and a preview chat let you try it before it goes live; n8n warns when a publish depends on unpublished workflow tools.
6. Connect the output. The agent’s result flows into the next nodes, such as writing to a CRM only after a human approves in Slack.

What It Does Alone and What Needs You

At run time the agent chooses tools and their order on its own, and refused tool calls now show the policy reason, which helps with debugging. Everything before that is human work: picking triggers, designing tools, writing prompts and deciding where approvals belong. The built-in n8n Assistant can draft workflows from a description and uses AI credits included in each cloud plan, but reviewers on Capterra still name the learning curve for people without API knowledge as the most common complaint.

Self-hosted users bring their own model keys, so model usage is billed by OpenAI, Anthropic, Google or whichever provider they choose, not by n8n. Cloud users can do the same, which keeps model choice open.

A Typical Agent Design

A common first project is a support triage agent. A Gmail trigger starts the workflow for each new message. The agent reads it, looks up the customer in HubSpot through a tool, checks order status in Postgres and drafts a reply. Instead of sending, the workflow posts the draft to a Slack channel and waits for a person to approve it, then sends and logs the result to Google Sheets. The agent supplies judgment in the middle; the fixed steps on either side keep it from acting where it should not. The same pattern works for lead qualification, invoice checks and internal help desks, and the budget caps added in October 2026 put a ceiling on what each one can spend on model calls.

n8n AI Agent Pricing

n8n charges by execution: one complete run of a workflow, however many steps it contains. That makes agent workflows with many tool calls cheaper than on step-based tools. Cloud prices are in euros.

The cloud trial includes 1,000 executions without a credit card, and a 14-day trial is available for Business features. Model usage is extra on every plan unless you use the included AI credits for building.

What It Costs in Practice

Small business support agent. An agent that triages about 80 support emails a day runs about 2,400 executions in a 30-day month, which fits inside Starter’s 2,500. That costs €240 a year, or about €0.008 per execution, plus whatever the model provider charges. A busy month will push past the limit, so watch usage.

Agency running client automations. At about 300 agent runs a day, a team needs roughly 9,000 executions a month, or 90% of the Pro allowance. Pro costs €600 a year, about €0.005 per execution, and adds admin roles and more concurrency. Because n8n counts runs rather than steps, an agent that makes ten tool calls per run still uses one execution.

Mid-size company that needs SSO. Business costs €667 a month, or €8,004 a year, for 40,000 executions, which is about €0.017 per execution, more than three times Pro’s rate. A company with fewer than 20 employees pays half, €333.50 a month. The alternative is the free Community Edition on your own server, which removes execution limits but adds hosting, maintenance and security work your team must own.

n8n AI Agent Compared With Zapier, Make and CrewAI

Buyers usually compare n8n with the two big no-code automation platforms and with code-first agent frameworks.

For an agent that makes many tool calls per run, n8n’s per-execution pricing is usually the cheapest of the hosted options, and independent comparisons put self-hosting on a small server at a fraction of Zapier’s cost for high volumes. Zapier wins on ease, and CrewAI on flexibility for developers.

Who Should Use n8n AI Agent

• Operations and RevOps engineers. Agents that read a ticket, look up the customer and draft a reply can sit next to fixed steps like CRM updates and approvals.
• Companies with data residency needs. Self-hosting keeps data on your own infrastructure, and the cloud version stores data in Frankfurt.
• Automation agencies. Unlimited users on cloud plans and per-run pricing make client work economical.
• Teams that want model choice. n8n works with OpenAI, Anthropic, Google and other providers, including Claude deployments through Azure AI Foundry.

Who Should Skip n8n AI Agent

• Non-technical teams. If nobody can debug an API call or design a prompt, a ready-made assistant such as Lindy will deliver value faster.
• Mid-size companies needing SSO on a small budget. The jump from €50 to €667 a month for SSO and version control is steep.
• Buyers who need a named certification on paper today. n8n offers its SOC 2 report to Enterprise customers; smaller plans get less documentation.
• Teams that cannot maintain servers. Self-hosting is free in license terms only; upgrades, backups and security are your job.

What to Watch Out For

• Breaking upgrades. n8n 2.0, released in late 2025, included breaking changes to security, storage and configuration, and reviewers say major updates have broken existing workflows. Test upgrades on a copy before production.
• Execution limits. Busy agents can exhaust Starter quickly, and concurrency limits of 5 on Starter and 20 on Pro cap how many runs happen at once.
• Model costs are separate. Set the agent budget and session spend caps added in October 2026, and set limits with your model provider too.
• Usage data. n8n’s AI terms rule out training on customer content, but allow aggregated, de-identified usage data and prompts to improve its AI services.
• Patching self-hosted installs. The Ni8mare vulnerability disclosed in January 2026 (CVE-2026-21858, rated 10.0) allowed full takeover of unpatched instances until version 1.121.0. Watch n8n’s release notes and upgrade promptly.
• Thin error messages. Reviewers mention vague errors and documentation gaps, which add time to debugging agent behavior.

Recent News

On May 12, 2026, SAP took a stake in n8n at a $5.2 billion valuation through a secondary share sale and signed a multi-year deal to make n8n available inside Joule Studio. In January, security firm Cyera disclosed a critical flaw nicknamed Ni8mare that allowed full takeover of unpatched instances; n8n fixed it in version 1.121.0.

n8n released version 3.0 on October 27, 2026, adding a Test agent button to the agent builder, mention-only Slack agents and new memory and vector store options, after a fall of agent-focused releases that brought monthly budget caps and Browser Use on October 6 and an agent setup checklist on October 13. The previous major version, n8n 2.0, arrived in November 2025 with breaking security changes, a redesigned canvas and a migration report tool for administrators.

Is n8n AI Agent Worth It? Our Verdict

For teams with someone technical on hand, yes. The n8n AI Agent scores 7.5/10 on our scale. Pricing value is its strongest point: a free self-hosted edition, low cloud entry prices and per-run billing that does not punish agents for making many tool calls. Integrations are deep, from any HTTP API to MCP servers and multiple model providers, and reviewers rate the platform 4.6 out of 5 on Capterra.

The score is held back by setup effort and autonomy that depends on what you build. n8n will not hand you a finished assistant, and upgrades can break workflows. Start on the free trial or Community Edition, build one agent for one process, and add budget caps before it touches customers. For more options see our workflow automation agents, agent builders and open-source agents pages.

n8n AI Agent FAQ

How Much Does n8n Cost?

The self-hosted Community Edition is free. Cloud Starter costs €20 a month and Pro €50 a month on annual billing, for 2,500 and 10,000 executions. Business costs €667 a month billed annually, with half off for companies under 20 employees, and Enterprise is quoted.

Is n8n Really Free?

The Community Edition has no license fee and no vendor execution cap, but you pay for hosting, model usage and your own time to maintain it. Business and Enterprise features on self-hosted installs need a paid license key.

Is n8n Safe for Company Data?

Cloud data is stored in Frankfurt, encrypted in transit and at rest, and n8n’s AI terms say customer content is not used to train models. Enterprise adds SSO, audit logs, log streaming and access to a SOC 2 report. Self-hosting keeps data entirely on your own infrastructure.

Is n8n Better Than Zapier for AI Agents?

For technical teams running many agent calls, usually yes, because n8n charges per run rather than per step and can be self-hosted. Zapier is easier and connects to more apps, which makes it the better choice for non-technical teams.

What Can’t the n8n AI Agent Do?

It does not come as a ready-made assistant; you have to design each agent’s triggers, tools and limits. It also cannot guarantee consistent behavior without testing, and it needs someone comfortable with APIs to troubleshoot.