n8n
Open-source workflow automation with a self-hostable engine, native AI nodes, and LLM-integrated workflow building.
n8n, pronounced n-eight-n, is an open source workflow automation platform with native AI capabilities, built for technical teams who want the speed of no code with the option to drop into real code. Developed under a source available fair-code license, it lets users connect hundreds of apps and services on a visual node based canvas, and write JavaScript or Python, add npm packages, or build custom nodes wherever a workflow needs logic that drag and drop cannot express. Workflows support branching, loops, error handling, and version control.
n8n can be fully self hosted, so a team runs automation on its own infrastructure, including air gapped deployments, and credentials and data never leave its environment. Pricing is based on workflow executions rather than per task or per operation, so a long, multi step AI workflow costs the same as a simple two step one.
n8n treats AI as part of the canvas rather than a bolt on action. Its AI Agent node, built on LangChain, creates autonomous agents that reason about a goal and decide which tools to call and in what order, assembled from modular sub-nodes for memory, tools, output parsing, and retrieval. Teams can build RAG pipelines against vector databases like Pinecone and Qdrant, route different steps to different model providers including OpenAI, Anthropic, Google, and local models through Ollama, and keep humans in the loop with approval steps, while every step of an agent's reasoning is traceable on the canvas.
Through native Model Context Protocol support, any n8n workflow can be exposed as a tool that an external AI model can call. For organizations, n8n adds SSO, role based access, encrypted secret storage, audit logs, and Git based controls.
Vendor details
Canonical URL
https://n8n.io/
Category
Agent builder
Subcategory
Fair-code workflow automation with LangChain-based agent nodes and self-hosting
Company status
independent
Use cases & customers
Primary use cases
Target customers
Deployment options
Integrations
Several hundred individually documented app nodes spanning CRM, marketing, ITSM, HR, databases and warehouses, cloud infrastructure, storage, messaging, security tooling, payments and finance, with major nodes carrying separate documentation per operation family. Core nodes add HTTP Request, GraphQL, SSH, Execute Command and Code for anything uncatalogued, and community and custom nodes written in TypeScript extend the catalog further. Tool Mode converts an ordinary node into an agent-callable tool, and sub-workflows can be attached to an agent as tools. Model Context Protocol runs both ways: the MCP Client node consumes external MCP servers, and the MCP Server Trigger node makes n8n itself an MCP server exposing its tools and workflows to MCP clients over SSE or streamable HTTP. Model providers available as first-class nodes include OpenAI, Anthropic, Google Gemini, Mistral, Perplexity, Qwen Cloud, MiniMax and Moonshot Kimi, plus local models through the self-hosted AI starter kit. Vector stores, embeddings, retrievers and document loaders attach as sub-nodes so the retrieval index can live in a customer-run database.
In practice
You can't send customer data through a third-party cloud for compliance reasons. n8n self-hosts on your own infrastructure, even air-gapped, so credentials and payloads never leave your environment while you still automate across hundreds of apps.
Your AI workflow has twenty steps and per task pricing would bankrupt it. n8n charges per execution, so a long multi step agent pipeline costs the same as a two step one.
You want an agent that triages support tickets by querying a database and drafting a reply, with a human approving before send. n8n's AI Agent node reasons over the goal, calls the tools, and pauses for approval.
Sources & related URLs
Related / legacy domains
Research sources
Agentic Index coverage score
13.0 / 14 capabilities · 93%
| Integrations & Tool Calling | Full |
|---|---|
|
Several hundred built in app nodes span CRM, marketing, ITSM, HR, databases, warehouses, cloud infrastructure, storage, messaging, security and finance, and major nodes such as Gmail, Google Drive, Google Sheets, Postgres and OpenAI each cover several families of operations. Core nodes add HTTP Request, GraphQL, SSH, Execute Command and Code for anything not in the catalog, community and custom nodes extend the set further, and the MCP Client node connects agents to external MCP servers. Sub-workflows and app nodes can be exposed to agents as tools. Sourcedocs.n8n.io app nodes, core nodes, MCP Client and how tools work pagesread 2026-08-31 |
|
| Workflow Orchestration | Full |
|
n8n workflows split with If and Switch conditionals, merge data, loop, wait, handle errors gracefully and run workflows with several branches in a defined execution order. Workflows can call other workflows, and converting nodes to a sub-workflow automatically brings along attached AI sub-nodes such as chat models and memory. Agent nodes act as root nodes orchestrating typed sub-node connections for language models, tools, memory and output parsers, and sub-workflows can be attached to an agent as tools. Canvas groups, sticky notes and workflow tags help organize large workflows. Sourcedocs.n8n.io flow logic tree, break workflows into smaller parts, convert to sub-workflows and how tools work pagesread 2026-08-31 |
|
| Knowledge Grounding & RAG | Full |
|
Data can be stored and searched with vectors through vector databases, embeddings, retrievers and document loaders, and retrieval augmented generation pulls relevant context into a workflow. Vector store providers attach as sub-nodes so the index can be held in a database the customer runs, and a worked example scrapes website content into Pinecone and queries it from a chat workflow. Data tables manage structured tabular data within project boundaries, and Default Data Loader and text splitter components handle ingestion. Sourcedocs.n8n.io store and search data with vectors, retrieve relevant context, use website content and data tables pagesread 2026-08-31 |
|
| Human Oversight & Guardrails | Full |
|
An AI Agent can be required to get a person's approval before it runs specific tools, and a human fallback brings in a person to answer when the AI cannot help. A dedicated Guardrails node ships as a core node for enforcing policy on AI workflows. Workflow reviews let a builder submit a version for review before publishing, compare changes in a visual diff, discuss them in activity comments, and approve or request changes. Wait and Stop And Error nodes pause and halt runs, and publishing a workflow is a separate step from saving it. Sourcedocs.n8n.io human-in-the-loop for tools, set a human fallback, Guardrails node, workflow reviews and save and publish pagesread 2026-08-31 |
|
| Security, Identity & Governance | Full |
|
An independent auditor evaluates n8n continuously and audits it each year against SOC 2. Its SOC 2 report is available to enterprise customers, a SOC 3 report is published for everyone else, and third party penetration tests of the production environment run at least once a year. SSO, SAML and LDAP come on the Enterprise plan and advanced RBAC on all paid plans. Security settings add SAML and OIDC single sign on, MFA enforcement, a self service security audit, node blocking, task runner hardening, SSRF protection, encryption key rotation, JWE decryption of OAuth tokens and execution data redaction. Sourcen8n.io legal security page, docs.n8n.io security configuration treeread 2026-09-29 |
|
| Observability & Auditability | Full |
|
Executions are first class objects. Teams can view all executions or those of a single workflow, add custom execution data so runs can be filtered by custom fields, and debug by copying earlier execution data into the canvas. Workflow and node executions can be traced to an OpenTelemetry collector, alongside Prometheus metrics, metrics visualized in Grafana, log configuration and an insights metrics layer. Workflow change history and version restore are available, execution data can be redacted for compliance, and a self hosted instance can connect to LangSmith for tracing through LangChain in n8n. Sourcedocs.n8n.io understand executions, trace executions with OpenTelemetry, monitor n8n, view change history and LangChain in n8n pagesread 2026-08-31 |
|
| Memory & State Persistence | Partial |
|
A memory sub-node on the Tools Agent lets users hold an ongoing conversation across several queries, but that memory does not persist between sessions, and the default Simple Memory node, previously named Window Buffer Memory, is a bounded buffer. Memory attaches to agents as a typed AiMemory sub-node, with session IDs taken from input data or a custom key and optionally scoped per node so separate memory nodes in one workflow do not share history. External chat memory backends can be attached in place of the simple buffer, and Data tables provide structured storage across runs. Sourcedocs.n8n.io Tools Agent, how memory works, AI Agent common issues and data tables pagesread 2026-08-31 |
|
| Deployment & Data Residency | Full |
|
n8n installs with a one line setup, Docker, Docker Compose or npm, and deploys to DigitalOcean, Heroku, Hetzner, AWS, Azure, Google Cloud Run, Google Kubernetes and OpenShift Local, alongside managed n8n Cloud. An isolated configuration keeps the instance from connecting with n8n's servers. The customer chooses the database, binary and execution data can go to external storage, queue mode and task runners cover scaling, and an OEM path puts n8n inside another product. Editions run from Community and Registered Community to Business and Enterprise. Sourcedocs.n8n.io install options, isolate n8n, choose n8n's database, scaling and compare editions pagesread 2026-08-31 |
|
| Prebuilt Agents, Templates & Packs | Full |
|
Templates are one of the named ways to build workflows, and self hosted instances can point at a custom workflow templates library. A curated self hosted AI Starter Kit provides a set of AI elements for starting AI workflows quickly. Each node links to browsable template collections for that node, and .n8np packages bundle workflows, folders or whole projects with their dependencies into a portable file for moving between instances. Learning paths offer interactive courses that build real workflows. Sourcedocs.n8n.io use templates, configure a custom workflow templates library, deploy with the AI starter kit and n8n packages pagesread 2026-08-31 |
|
| Triggers & Channel Coverage | Full |
|
Trigger nodes include Schedule Trigger, Webhook, Chat Trigger, Email Trigger over IMAP, RSS Feed Trigger, Local File Trigger, SSE Trigger, n8n Form Trigger, Error Trigger, Activation Trigger, Workflow Trigger, Execute Sub-workflow Trigger, n8n Trigger, Evaluation Trigger, MCP Server Trigger and Microsoft Agent 365 Trigger. On self hosted instances a durable scheduler runs time based workflows from a queue backed by the database. Agents can be published so people reach them through chat, channels and schedules, and a Respond to Webhook node plus streaming responses complete the request and response surface. Sourcedocs.n8n.io core node reference, durable scheduler, build and manage agents and stream real-time responses pagesread 2026-08-31 |
|
| Model Flexibility & Routing | Full |
|
The AI Agent node needs a connected Chat Model sub-node and errors without one, and clicking the Chat Model connector opens a selection of possible chat models. Models attach as sub-nodes, so different agents and chains in one workflow can use different models. Dedicated model provider nodes include OpenAI, Anthropic, Google Gemini, Mistral AI, Perplexity, Qwen Cloud, MiniMax and Moonshot Kimi, and the self hosted AI starter kit runs local models alongside n8n. LangChain is the underlying abstraction, with a LangSmith connection. Sourcedocs.n8n.io AI Agent node and common issues, what agents do, app node reference and AI starter kitread 2026-08-31 |
|
| APIs, SDKs & MCP Extensibility | Full |
|
The MCP Server Trigger node lets n8n act as a Model Context Protocol server, making n8n tools and workflows available to MCP clients over SSE or streamable HTTP. MCP clients can build and run n8n workflows, and admins oversee connected MCP clients and can revoke their consent. A public REST API is available and can be disabled. Custom and community nodes extend the catalog, npm modules can be enabled in the Code node, and .n8np packages export workflows, folders or whole projects as portable files with a manifest, size limits and API key scopes. An OEM path embeds n8n in another product using OAuth 2.0 token exchange, iframe SSO and delegated API access. Sourcedocs.n8n.io MCP Server Trigger, connect to n8n MCP server, n8n packages, security and OEM deployment pagesread 2026-08-31 |
|
| Testing, Debugging & Optimization | Full |
|
Evaluation and Evaluation Trigger core nodes ship with the platform. Builders run quick evaluations during development against known test cases and use metric based evaluations to measure, score and improve production workflow performance over time, with guidance on why to test and how to fix common issues. Debugging uses pin and mock data, debug executions that copy earlier execution data into the current workflow, manual and partial runs, and dirty node tracking. Sourcedocs.n8n.io test and improve AI workflows section, Evaluation and Evaluation Trigger nodes, understand executions pagesread 2026-08-31 |
|
| Browser & Computer Use | Partial |
|
n8n ships an Airtop app node that controls a cloud based web browser for querying, scraping and interacting with web pages, with operations to manage browser sessions and windows and to click an element, fill a form, hover, scroll and type, authenticated with Airtop credentials. Airtop is a third party browser engine wired in through a node, not a browser n8n ships itself. None of n8n's core nodes is a browser, desktop or computer use node, and SSH and Execute Command run commands on remote hosts. Sourcedocs.n8n.io Airtop node pageread 2026-09-29 |
|
The Agentic Index coverage score grades every vendor Full, Partial or Not documented against the same 14 buyer facing capabilities, from public evidence only. Each capability links to how all vendors in the index score on it. How this evidence is graded
Product updates
Version 3.0 is out as a public beta, a major release that clears out legacy features and makes security defaults stricter, with the stable version due in early December. Self hosted n8n now runs only on Docker, as npm installs are no longer supported, while Kubernetes and Helm setups are unaffected. Slack credentials now need a signing secret, unverified community packages are off by default, and the default Code node timeout drops from 5 minutes to 1.
Bears on: Security / enterprise
View sourcen8n 2.43 makes Browser Use in the n8n Assistant available to all users, where it had been limited to an experiment. Instance admins control it with a single environment setting, which is on by default.
Bears on: Agent capability
View sourceVersion 2.42.0 switches on n8n's Agents builder by default for self hosted installs and adds approval settings for agents working in Slack, Telegram, Discord and Linear. A connection can require approval before risky actions or before every action, and the agent then pauses with an Approve or Deny card in the channel before it posts or files anything.
Bears on: Agent capability
View sourcePricing
Starter from €20 a month billed annually (€24 monthly), Pro €50 (€60 monthly), Business €667 (€800 monthly) and Enterprise custom. The self hosted Community Edition is free.
Tiered subscriptions priced in euros, each with a monthly allowance of workflow executions. One full run of a workflow counts as one execution, however many steps it has.
Included quota
The self hosted Community Edition is free, with unlimited executions, workflows and users and all integrations, on a server the customer supplies. Cloud Starter (€20 a month billed annually) includes 2,500 executions a month, 5 concurrent executions and unlimited workflows. Cloud Pro (€50 a month billed annually) includes 10,000 executions a month, 20 concurrent executions, admin roles, global variables and 7 day insights. Business (€667 a month billed annually, self hosted, for companies under 100 employees) includes 40,000 executions a month, 6 projects, SSO (SAML and LDAP), Git version control, multiple environments and 30 day insights. Enterprise has unlimited executions, advanced roles, dedicated support and custom deployment.
What is public
n8n (n8n.io, from Berlin based n8n GmbH) sells fair-code workflow automation with visual and code building and native AI. It prices in euros and bills by executions, where one full workflow run is one execution regardless of node count. The self hosted Community Edition is free forever, with unlimited executions, workflows and users. Cloud, hosted in the EU in Frankfurt, offers Starter at €20 a month billed annually (€24 monthly, 2,500 executions) and Pro at €50 a month billed annually (€60 monthly, 10,000 executions). Self hosted Business is €667 a month billed annually (€800 monthly, 40,000 executions, SSO and Git), and Enterprise is custom with unlimited executions. Every Cloud plan has a 14 day free trial, and there is no permanent free Cloud tier. A Start-up Plan gives 50% off Business to companies under 20 employees.
Billing mechanics
Billed per execution. A single end to end run of a workflow counts as one execution no matter how many steps or nodes it has or how much data it moves, so a 2 step and a 50 step workflow each use one execution per run. Only successful executions count, and failed and test runs are free. No plan limits active workflows, so the charge rests on executions alone. Annual billing saves 17%. Business is self hosted under a license key, and Business and Enterprise features check that key with n8n's license server daily.
Cost watchouts
Execution caps bite on polling and scheduled workflows. A single workflow that runs every 5 minutes generates about 8,640 executions a month, which uses up the 2,500 Starter cap in about 9 days. Cloud has no permanent free tier, and Starter and Pro have no overage, so workflows pause hard at the cap. The 17% annual discount comes with no mid year refund. Cloud limits stored credentials. Self hosted software is free, but the customer owns the server and its maintenance.
Variable cost rationale
Executions are metered. Each full workflow run counts as one execution however many steps it has, and only successful runs count. The bill grows with how often workflows run. Schedules and polling triggers add executions fast, and a workflow that runs every 5 minutes makes about 8,640 executions a month. Starter and Pro cap executions each month at 2,500 and 10,000, and workflows pause at the cap instead of running into overage. Business buys extra buckets of 300,000 executions for €4,000, and Enterprise is unlimited. Self hosting the free Community Edition turns the cost into server infrastructure the customer runs.
Additional watchouts
Polling and scheduled workflows can exhaust execution caps fast, and Starter and Pro pause with no overage. There is no permanent free Cloud tier. Prices are in euros. Self hosting needs server and DevOps work, including updates, SSL and security.
Overage / add-ons
On Starter and Pro, workflows pause when they reach the monthly execution cap and resume the next cycle, with no overage option. On Business, extra executions are bought in buckets, with 300,000 more executions for €4,000. Enterprise has unlimited executions, and the self hosted Community Edition has no caps.
Sales call required
Mixed (some tiers require a call)
Free / trial
Free self hosted Community Edition, and a 14 day Cloud trial with no card required (the Business trial needs a card)
Lowest paid plan
Starter at €20 per month billed annually, or €24 billed monthly
Commercial notes
Fair-code and source available, with AI agent and workflow features, 400+ official integrations, and extensions through code and community nodes. Self hosting runs on infrastructure such as Hetzner or Coolify, or on managed hosts like PikaPods and InstaPods. n8n Cloud runs on servers in Frankfurt and is GDPR compliant, and self hosting covers other regions or full data control.
Key ambiguities
n8n prices in euros, so dollar equivalents vary with exchange rates. There is no free Cloud tier, self hosted Enterprise is tiered, and only successful executions count. n8n notes that plans and allowances change often. The Business plan is self hosted only.
Cancellation / refund
Monthly plans cancel at any time and end with the billing cycle, annual plans run to the end of the term, and there are no pro rata refunds. Annual billing is 17% cheaper.
Support SLA / resale
The Community Edition, Starter and Business get community forum support. Enterprise has dedicated support with an SLA.
Missing data
No dollar prices are published, since n8n prices in euros. Plans, allowances and AI credit grants change often.
Related vendors
- Lindy — Lindy is an AI teammate that lives in Slack, answers from your tools…
- Activepieces — Open source, MIT licensed automation platform where every one of 700…
- Agenta — Agenta is an open source workspace where teams build AI agents by…
- AgentX — No code platform to build, evaluate, and deploy multi agent…
- AI Library — No code platform for building and running AI agents, with a…
- Aigensei — Enterprise agentic workflow platform that runs business processes…
Alternatives to n8n
The closest documented capability profiles to n8n among agent builders tracked by Agentic Index, ordered by similarity on the same 14 point evidence the rankings use. No vendor pays for placement.
- ZBrain13.0 / 14Matches n8n across all 14 documented capabilities
- Agenta12.5 / 14A lighter documented profile than n8n
- Gumloop13.5 / 14Fuller documented coverage on Browser & Computer Usen8n vs Gumloop →
- Mendix12.5 / 14A lighter documented profile than n8n
- OutSystems12.5 / 14A lighter documented profile than n8n
- StackAI13.5 / 14Fuller documented coverage on Browser & Computer Use
Similarity is computed from each vendor's Agentic Index coverage score evidence, axis by axis, not from the totals. How this evidence is graded