n8n AI Agents
n8n is a workflow-automation platform for technical operations teams that need to combine AI agents with deterministic steps, integrations, code, and self-hosted infrastructure. Technical teams can connect ecommerce systems and APIs, choose cloud hosting or operate their own instance, and add JavaScript or Python when visual nodes are not enough. Self-hosting gives the operator more control over the runtime and data path, but also transfers responsibility for credentials, security, monitoring, backups, and upgrades.
n8n is a workflow-automation platform for technical operations teams that need to combine AI agents with deterministic steps, integrations, code, and self-hosted infrastructure. Technical teams can connect ecommerce systems and APIs, choose cloud hosting or operate their own instance, and add JavaScript or Python when visual nodes are not enough. Self-hosting gives the operator more control over the runtime and data path, but also transfers responsibility for credentials, security, monitoring, backups, and upgrades.
n8n is a workflow-automation platform for technical operations teams that need to combine AI agents with deterministic steps, integrations, code, and self-hosted infrastructure.
- Views
- 0
- Uses
- 0
- Favorites
- 0
- Supported Platforms
- Shopify · Salesforce · Slack · Google Drive · HubSpot · GitHub · OpenAI · Anthropic · API integrations
Buying view: n8n AI Agents fits technical teams that want agent steps inside a broader workflow-automation system and need a choice between cloud and self-hosted operation. Self-hosting increases control over execution and data paths, but the operator then owns credentials, security, monitoring, backups, model costs, and upgrades.
Capabilities
n8n helps a team connect store work across systems: answer policy questions from approved documents, or turn a new order into an email draft for a colleague to review. Operations staff decide what information to use, which rules to follow and who checks the result. A technical colleague connects the apps, AI model and processing steps on a visual canvas, then tests and publishes the workflow.
Choose how people and events start the work
A shopper asking a question and a store receiving an order need different starting points. n8n can start with a chat message or a business event, and use an Agent where the task calls for choosing among connected tools. A team can combine these approaches in one workflow.
Conversational assistants
A shopper or colleague sends a message through a hosted or embedded chat. Chat Trigger passes it to an Agent or chain and returns the configured response. The builder chooses access settings and whether to connect memory for conversation history.
Chat Trigger ↗Event-driven workflows
An app event, schedule or webhook starts a defined sequence of work. The team connects data, conditions and actions on the canvas; selected steps can use AI to summarize or draft text. Publishing enables the configured production triggers.
Save and publish workflows ↗Agents with selected tools
An AI Agent can choose among the tools connected by the builder and use their results to continue a task. The team defines instructions, credentials and limits. A configured tool is an available action, not a step that runs on every request.
Tools AI Agent ↗Operations staff own the business rules and source material; a technical colleague handles app access, setup and failures. Both roles are needed to keep a published workflow useful as store policies and connected systems change.
Combine the capabilities the task needs
A policy assistant needs a question, relevant store documents and a reply. An order follow-up needs an order event, a selection rule and a mailbox to save the email draft. The builder connects only the components needed for that result and checks what each step passes to the next.
Messages and business events
Choose an entry that matches the work: a chat message, a scheduled run or an event from a connected app. Test the incoming fields before passing them on; missing data should remain missing rather than being guessed by AI.
Create and run workflows ↗Models and tool selection
An Agent can read a request, choose a connected tool and use the returned information to continue. For a policy question, that tool might search the store documents before the model writes a reply. The builder describes each tool, connects a supported chat model and limits how many times the Agent can work through the task.
Tools AI Agent ↗Answers grounded in store documents
A policy assistant can look up the relevant delivery or returns passage before answering a shopper. The builder splits the documents into passages and uses an embedding model to make them searchable in a vector store, a database used for retrieval. Document loading and passage retrieval based on a question must use the same embedding model; operations keeps the source policies current.
Retrieve relevant context ↗Explicit conditions and branches
The team can state a rule such as “prepare a follow-up only when the order has a usable customer email address.” An If node separates orders that meet that condition from those that need a colleague to check the missing information. A Switch node can route work to more than two paths when the task has several categories.
If and Switch branches ↗Actions in connected accounts
A workflow can put its result where the next colleague will use it. For example, Gmail’s Draft operation saves generated text as an email draft, ready for an operator to check and send. The builder connects the mailbox account and maps the recipient, subject and body to that operation.
Create Gmail drafts ↗Approval before selected Agent tools
Attach human review to tools that need oversight and configure a review channel. n8n pauses that tool call for approval; denial cancels it and informs the Agent. Explain denial handling in the Agent instructions and define how the team follows up on unanswered requests.
Human review for tools ↗From a workflow idea to live operation
Operations staff first agree on the task, sample inputs and expected result with the builder. The team tests that result before publishing, then uses saved execution records to investigate problems and refine the workflow. Later edits remain in the draft until the revised version is published.
- Choose entry
Create a workflow and choose the trigger that matches the intended task and users.
- Connect steps
Configure models, tools and credentials, then connect the required business conditions and outputs.
- Test inputs
Run representative inputs manually and inspect the intermediate results before making the workflow live.
- Publish workflow
Publish the intended version to enable its configured production triggers; check the public chat settings separately when applicable.
- Inspect runs
Review saved execution details, diagnose failures and revise the workflow when the task or data changes.
A store-policy assistant for shoppers
A shopper asks how to return an unopened item. The assistant searches the store’s current policy and explains the next step in chat, so staff can focus on requests that need an individual check. This example covers public policy questions; private order lookups and refunds are not connected.
Before building
For this illustrative store, the policy says: “For an unopened item, contact customer service with the order number to request a return.” Operations prepares that policy, the support contact and sample questions; this is an example rule, not a rule supplied by n8n. A technical colleague arranges the n8n and model accounts, an embedding model for document retrieval and a vector store that retains the imported documents.
Build and test
Prepare searchable policy documents
Operations confirms that the returns document contains the example rule and the current support contact, and removes private customer details. The builder loads the document, splits it into searchable passages, processes them with the embedding model and stores them in the vector store. Before connecting chat, they check that a search about unopened returns retrieves the intended passage.
Connect the chat and Agent
The builder adds Chat Trigger, an AI Agent and a supported chat model. The chat stays private during setup. The Agent instructions define the assistant's role and require it to use the connected retrieval tool when answering policy questions rather than invent store rules.
Connect retrieval and the response
The vector store becomes an Agent tool with a clear description of its policy content. The builder uses the same embedding model as the loading workflow, checks the returned passages and returns the Agent output through Chat Trigger. If history is needed, both chat and Agent connect to the same memory component.
Define the unsupported-question response
Operations defines a fallback such as “The available policy does not cover this situation; please contact customer service,” followed by the store’s contact details. The builder adds it to the instructions and tests questions about opened items and private orders. This gives the shopper a next step; a separate integration would be needed to create a support ticket.
Test, publish and maintain
The team tests a question covered by the documents, a question not covered by them, a failed retrieval connection and a request for another shopper's order. The builder addresses any issues found during testing, publishes the workflow and configures hosted or embedded public chat. A technical colleague handles embedding the chat on a webpage; operations updates the source documents and checks answers after each reload.
Illustrative run
Illustrative run: the shopper asks, “How do I return an unopened item?” After retrieving the example policy, the assistant replies, “Please contact customer service with your order number to request a return,” and provides the configured contact. If the shopper asks whether an opened item qualifies, the assistant directs them to customer service because the example document does not answer that question.
What the team receives
The shopper receives the documented next step without waiting for staff to repeat a standard policy answer. Operations reviews unanswered questions and updates the source documents when policies change. The team also samples generated replies to check that they stay within the retrieved policy; retrieval does not guarantee a correct answer every time.
Prepare follow-up drafts for new Shopify orders
After a new Shopify order arrives, this example prepares a thank-you email when a usable customer email address is available. AI drafts the message from the supplied order facts, and Gmail saves it for the mailbox owner to review. Sending remains a manual step, so the colleague can check the wording and avoid repeating an existing store notification.
Before building
Operations sets the example rule: prepare a draft only when the new order includes a usable customer email address; otherwise leave it for manual investigation. The message should thank the shopper and include the order reference, without adding an unverified delivery promise. An administrator connects Shopify, Gmail and a model account; a technical colleague also sets up a lasting record of processed orders to prevent repeated notifications from producing repeated drafts.
Build and test
Receive the order event
The builder selects Order Created in Shopify Trigger and checks a sample event from the connected store. They locate the order ID, the order reference used in the subject and the customer email address, then check that the values reach the next step correctly. These values come from the event data, not from the language model.
Prevent duplicate drafts
The workflow checks whether the order is already being handled or already has a saved draft. If it is new, it reserves the order for processing before continuing; a technical colleague must make this check safe when two notifications arrive together. If the record cannot be read or the last attempt has an uncertain result, processing stops for investigation before another Gmail draft is created.
Apply the business condition
The builder checks that the customer email address is present and usable, then passes that result to an If node. Orders that pass continue to drafting; those that fail leave this drafting path for a colleague to investigate. This is the example store’s follow-up rule, and it does not change the order or pause fulfillment.
Generate and save a draft
The builder asks an AI text-generation step to write a brief thank-you using the supplied order reference, without inventing shipping dates or discounts. Gmail’s Draft → Create operation receives the generated text; its recipient is mapped directly from the checked email address and its subject includes the order reference. After Gmail confirms creation, the workflow saves the returned draft ID beside the order ID for later checks.
Test the exits and enable the workflow
The team tests matching and non-matching orders, missing fields, repeated events and failed Gmail access. A colleague also checks whether the draft preserves the supplied facts. The builder publishes only after addressing any issues found during testing; the mailbox owner then checks drafts routinely and sends or discards them manually.
Illustrative run
Illustrative run: order #1042 arrives with a usable customer email address. The workflow finds no earlier processing record, checks the email and saves a Gmail draft with a subject such as “Thank you for order #1042” and a short thank-you message. The mailbox owner checks and sends it manually; a repeated notification for #1042 finds the processing record and produces no second draft.
What the team receives
The mailbox owner receives a draft with the customer address, order reference and prepared message, ready to review alongside the order. The workflow stops at saving the draft; it does not send the email or change the Shopify order. Preventing duplicate drafts depends on the tracking logic the technical colleague implements and tests.
Commercial plans
View official pricing detailsn8n's public Pro selector changes execution capacity and related limits. Cloud Starter/Pro and self-hosted Business belong to different deployment paths, not one interchangeable upgrade ladder. The prices below refer to the stated baseline volume; model-provider charges and self-hosted infrastructure are separate. Use the official selector for other capacities.
n8n public plans
| Product or sales surface | Scope | Public price and status |
|---|---|---|
| Starter | Hosted entry plan priced by workflow executions, not workflow steps. Included plan entitlements
| €20/month billed annually |
| Pro | Hosted plan for solo builders and small production teams. Included plan entitlements
| €50/month billed annually |
| Business | Self-hosted collaboration and scale plan for companies with fewer than 100 employees. Included plan entitlements
| €667/month billed annually |
| Enterprise | Hosted or self-hosted plan for strict governance and support requirements. Included plan entitlements
| Custom quote |
| Community | Open-source self-hosted edition for operators managing their own infrastructure. Included plan entitlements
| Free software; infrastructure not included |
Comparable tools: price and workflow
| Tool | Workflow difference | Official public price reference |
|---|---|---|
| Make AI Agents | Make AI Agents: Build visual AI agent workflows across 3,000+ apps with drag-and-drop simplicity. n8n AI Agents: Build self-hosted AI agent workflows with visual nodes, code, and API integrations. | Free — $0/month; up to 1,000 credits/monthCore — $12/month for 10,000 creditsPro — $21/month for 10,000 creditsTeams — $38/month for 10,000 creditsEnterprise — custom pricingMake official pricing ↗ |
| Zapier Agents | Zapier Agents: Goal-driven AI agents that browse, use knowledge, and take actions through Zapier connections. n8n AI Agents: Build self-hosted AI agent workflows with visual nodes, code, and API integrations. | Agents Free — $0; 400 activities/monthAgents Pro — $33.33/month, billed $400 annually; 1,500 activities/monthAgents Enterprise — contact sales; custom activity allowanceZapier Agents official pricing ↗ |
| Botpress | Botpress: Visual and code-extensible AI agent builder for custom support and commerce workflows. n8n AI Agents: Build self-hosted AI agent workflows with visual nodes, code, and API integrations. | Free — $0; 25 conversationsPlus — $150/month billed annually; 250 conversations/monthTeam — $750/month billed annually; 1,500 conversations/monthEnterprise — custom pricing and conversation volumeBotpress official pricing ↗ |
| Dify | Dify: Self-hostable visual LLM application platform for RAG chatbots, agents, and workflows. n8n AI Agents: Build self-hosted AI agent workflows with visual nodes, code, and API integrations. | Sandbox — freeProfessional — $59/workspace/month or $590/yearTeam — $159/workspace/month or $1,590/yearCommunity self-hosted — free software; infrastructure excludedEnterprise self-hosted — custom quoteDify official pricing ↗ |
| ClawTeams | ClawTeams: Delegate ecommerce operating goals to a coordinated AI team from supported chat tools. n8n AI Agents: Build self-hosted AI agent workflows with visual nodes, code, and API integrations. | First goal — $0 one timePaid subscription — no numeric public price displayedClawTeams official product page ↗ |
Frequently asked questions
What are n8n AI agents, and how do they work?
n8n supports agents in two related forms. In a workflow, an AI Agent node uses a connected chat model and at least one tool, then decides which available tool to call for the task. The separate Agent Builder creates agents alongside workflows with configured instructions, tools, knowledge, memory, channels, and schedules; it is currently in Preview. In both forms, an agent can use only the capabilities and credentials that the builder attaches.
Can a team build an n8n AI agent without coding?
A basic agent can be assembled with n8n's visual controls. For a chat workflow, the builder connects a Chat Trigger to an AI Agent node, adds a chat model and tools, writes the system instructions, and adds memory when later messages need earlier context. In Agent Builder, the team creates an agent from the project's Agents tab, configures it, tests the draft in Preview, and publishes a version. Custom APIs, data transformations, and production security can still require technical work.
Are n8n AI agents free to use?
Workflow-based agents can run on the free self-hosted Community edition, but the operator still pays for infrastructure, model inference, and any paid third-party tools. n8n Cloud becomes paid after its trial and uses plan execution quotas. Each turn with an Agent Builder agent counts as one execution and shares the workflow quota; model calls may use the team's provider account or eligible Gateway credits. These costs are separate from n8n Assistant credits used while building.
What can an ecommerce team automate with n8n AI agents?
An agent can interpret an open-ended request, retrieve information, and choose among attached tools for work such as support triage, order lookup, document questions, research, or updates to connected systems. For example, a support workflow can retrieve an order, search approved help content, and draft a reply. The required apps, permissions, and data sources must be connected first, and sensitive sends or record changes can be routed through human approval.
Where can a team find n8n AI agent workflow templates?
n8n's official workflow library has an AI category with templates for chat agents, document question answering, research, and other agent patterns. Before using a template, the team should review its required services, replace sample credentials, prompts, and data sources, then test every action. A free template does not make its hosting, model calls, or connected services free.
Which tools and apps can an n8n AI agent use?
A workflow-based agent can use supported app nodes, API requests, other n8n workflows, and MCP tools that are connected through its tool port. This can include reading a spreadsheet, querying a database, updating a CRM, calling an API with the HTTP Request Tool, or handing a multi-step task to another workflow. The agent chooses only among the tools made available to it, and every action remains limited by the attached credentials and permissions.
Can n8n AI agents answer questions from a team's own documents with RAG?
Yes. A workflow-based setup can load documents, split them into smaller passages, create embeddings, store them in a vector database, and connect that store as an agent tool. Retrieval-augmented generation, or RAG, gives the model relevant document passages at answer time; it does not retrain the model. Conversation memory serves a different purpose by retaining earlier messages, and important answers still need source and accuracy checks.
Can n8n AI agents be self-hosted with local AI models?
Yes for workflow-based agents. A self-hosted n8n instance can connect an AI Agent node to Ollama through the Ollama Chat Model node when the selected model supports tool calling. n8n's Self-hosted AI Starter Kit combines n8n, Ollama, Qdrant, and PostgreSQL as a testing foundation, but n8n says it needs security and production hardening before live use. Keeping n8n local does not keep data local when cloud models or external tools are still connected.
Native connections
Use the built-in node to read and manage products and orders in authorized Shopify stores; operations outside the node require a separate API route.
n8n Shopify nodeCreate, retrieve, update, and delete WooCommerce customers, orders, and products with store credentials configured in n8n.
n8n WooCommerce nodeRead, label, draft, reply to, or send email from an authorized mailbox; selected sends can pause for approval.
n8n Gmail nodeLet workflows read Slack data and post or update messages in the workspace allowed by the connected Slack credential.
n8n Slack nodeRead, append, and update spreadsheet rows, so workflows can use a sheet as an operating list or write results back for review.
n8n Google Sheets nodeConnect OpenAI credentials for model and media operations, including text, image, audio, file, and conversation tasks supported by the node.
n8n OpenAI nodeUse an Anthropic chat model as the model attached to an n8n AI chain or Agent; model access and usage charges remain with the configured Anthropic account.
n8n Anthropic Chat Model nodeSelect, insert, update, or delete database rows, or run a query, within the permissions of the configured database credential.
n8n Postgres nodeUse the built-in Stripe node for supported customer, charge, coupon, source, and token operations with an authorized Stripe account.
n8n Stripe nodeCreate and manage supported Zendesk tickets, users, and organizations inside a workflow using authorized Zendesk credentials.
n8n Zendesk nodeSources
- n8n AI Agents official product page
- GitHub product-scoped review collection
- Capterra product-scoped review collection
- G2 product-scoped review collection
- n8n AI Agent node documentation
- Official pricing and purchasing conditions
- Shopify connection
- Official workflow guide 1: Let an agent choose tools within a workflow
- Chat Trigger
- Tools AI Agent
- Retrieve relevant context
- Human review for tools
- Create and run workflows
- Save and publish workflows
- Inspect workflow executions
- If and Switch branches
- Shopify Trigger
- Create Gmail drafts
- Basic LLM Chain
- Make official pricing
- Zapier Agents official pricing
- Botpress official pricing
- Dify official pricing
- ClawTeams official product page
- n8n WooCommerce node
- n8n Gmail node
- n8n Slack node
- n8n Google Sheets node
- n8n OpenAI node
- n8n Anthropic Chat Model node
- n8n Postgres node
- n8n Stripe node
- n8n Zendesk node
- Browse all n8n integrations
- n8n Agent Builder
- n8n plans and editions
- n8n Gateway credits
- n8n AI workflow templates
- Using n8n workflow templates
- How n8n AI tools work
- n8n RAG guide
- How n8n AI memory works
- n8n Ollama Chat Model node
- n8n Self-hosted AI Starter Kit
User reviews
74 reviewsChecked Sep 20, 2026
Community reviews of n8n as a whole, including cloud and self-hosted use. The cards below are editorial summaries.
Review summary (our paraphrase)
Review summary (our paraphrase)
Reported a demanding setup and learning curve, followed by reliable self-hosted automation once the system and node logic were understood.
Read original reviewReview summary (our paraphrase)
Highlighted the source-available self-hosted option as a way to keep client data under the operator's control and reduce license costs.
Read original review314 reviewsChecked Sep 20, 2026
Reviews of the n8n workflow-automation product. G2's current review count was verified, but no complete review body was captured for a public card.
Individual reviews cannot be linked here. Open the original platform to read them.
More reviews on G249 reviewsChecked Sep 20, 2026
Company-profile reviews that may discuss n8n Cloud, self-hosting, billing, or support. The current count was verified, but stable individual-review links were unavailable for public cards.
Individual reviews cannot be linked here. Open the original platform to read them.
More reviews on Trustpilotn8n
Content checked:
Found visual workflows easier to hand over than scattered scripts and valued self-hosting, traceable flows, and extensible nodes.
Read original review