n8n introduced n8n Assistant on September 9, 2026 as a preview for building and troubleshooting automations through chat. Describe an outcome, and the Assistant can plan a workflow, add and connect nodes on the normal n8n canvas, request the connections it needs, run a test, inspect errors, and propose fixes.
The important part is what remains after the chat: a standard, visible n8n workflow that a teammate can inspect and edit. That makes the tool useful for getting past a blank canvas, but it does not make review optional. n8n explicitly says generated workflows may need correction and should be checked before production use.
What is n8n Assistant?
n8n Assistant is a chat-based workflow agent that can create, edit, test, and troubleshoot n8n automations from natural-language instructions. It builds ordinary workflows on the n8n canvas, uses the signed-in user's permissions, prompts for credentials through n8n's protected credential interface, and asks for confirmation before high-impact actions such as publishing or deleting.
It supersedes n8n's earlier AI Workflow Builder. The practical difference described by n8n is that the new Assistant does more than generate a first draft: it can execute the workflow, read node-level results, identify failures, apply an agreed fix, and run the workflow again.
n8n Assistant availability and cost
| Deployment or plan | Current preview status |
|---|---|
| n8n Cloud Starter | Available |
| n8n Cloud Pro | Available |
| Self-hosted Community | Available with supported setup |
| Self-hosted Registered Community | Available with supported setup |
| Self-hosted Business | Available with supported setup |
| Cloud or self-hosted Enterprise | Not generally ready; contact n8n about preview access |
| Self-hosted npm installation | Not supported in the launch guidance |
n8n says Docker-based self-hosting requires a supported version from 2.36 onward plus additional environment configuration and your own model keys. Check the current setup documentation before upgrading because preview requirements can change.
Assistant usage consumes AI credits according to tokens processed by the underlying model. Larger workflows, long conversations, and repeated debugging loops use more credits. The current pricing page lists 2,300 monthly AI credits for Starter and up to 13,700 for Pro, but plan allocations can change. Treat the live pricing page and your instance's usage display as the source of truth.
What n8n Assistant can do
- Create a workflow: turn a stated trigger, services, data rules, and desired outcome into connected nodes.
- Edit an existing workflow: add steps, change logic, or adjust node configuration through conversation.
- Test and troubleshoot: execute a workflow, inspect relevant node inputs, outputs, and errors, then suggest or apply a fix.
- Build agents: suggest instructions, tools, and skills for an agent where the deployment supports n8n's agents module.
- Connect MCP servers: use approved servers from n8n's registry as tools or context sources.
- Create supporting resources: work with resources such as Data Tables when the automation needs them.
The Assistant is not a background operator that studies an instance and invents automations without being asked. At launch it works within one instance, does not proactively monitor workflows, and does not drive a user's computer or browser.
A prompt that gives the Assistant enough detail
A request such as “automate invoices” leaves too many decisions unstated. A stronger first prompt identifies the trigger, systems, data, success path, failure path, test method, and approval boundary:
Create a draft workflow that runs every weekday at 8:00 AM.
Read new invoice emails from the Accounts Payable Gmail label,
accept PDF attachments only, save them to the correct vendor folder
in Google Drive, and add filename, sender, date, and Drive URL to an
n8n Data Table. If a required field is missing, send the item to a
manual-review branch. If any step fails, notify the finance Slack
channel without including invoice contents. Test with sample data,
show me the execution results, and ask before publishing or activating.This prompt still leaves credentials to n8n's connection flow. Never paste passwords, API keys, session cookies, or private invoice data into the chat merely to make a test easier.
How to build a workflow safely
- Start with a narrow outcome. Choose one recurring process with a measurable finish, rather than asking for a department-wide automation.
- Name every boundary. State which records may be read or changed, what should be excluded, and which actions require confirmation.
- Use sample data first. Test with realistic but non-sensitive examples, including an empty field, a duplicate, an API error, and an unexpected file type.
- Inspect the canvas. Open each node and verify filters, mappings, credentials, expressions, retry behavior, and destinations.
- Read execution data. A green final node can hide silently dropped items. Compare input and output counts and inspect each branch.
- Add an explicit failure path. Decide whether to retry, pause, alert a person, or place the record in a review queue.
- Publish with limited scope. Begin with a small data set or test account, watch early executions, and keep a rollback path.
When a transformation fails because incoming data is malformed, ToolNexa's JSON formatting and validation guide covers common syntax and structure problems worth checking before changing the entire workflow.
Credentials, permissions, and web access
n8n's documentation says credential secrets are not sent to the AI and are not visible to it. When a node needs access, the Assistant presents a credential card so the user can select or create a credential through the normal protected screen. That is safer than putting a secret into a prompt, but it does not remove the need for least-privilege accounts.
The Assistant inherits the signed-in user's access to projects, workflows, credentials, and supporting resources. Use a user and service accounts with only the permissions required for the workflow. A powerful workspace role can also give an agent a larger action surface.
If web access is enabled, n8n says the Assistant asks for approval before accessing an external domain. Check the exact domain and business need before approving it. Workflow structure, prompts, selected execution details, credential names or types, connection status, and approved web content may be processed to perform the task, so remove sensitive payloads that are not necessary.
Five production checks the preview cannot replace
- Business logic: confirm that every filter reflects the actual policy, especially dates, currencies, status values, and exclusions.
- Idempotency: make sure retries do not create duplicate tickets, payments, messages, or database rows.
- Error handling: test rate limits, expired credentials, timeouts, partial responses, and unavailable services.
- Data exposure: inspect logs, chat context, notifications, and third-party destinations for information that should be redacted.
- Ownership: assign a person to review failed runs, update connections, and approve future workflow changes.
The safest interpretation of “the workflow ran” is that one tested path completed. It is not proof that every edge case is correct. Use deterministic checks where possible, including field validation, count comparisons, test accounts, and approval nodes before consequential actions.
Frequently asked questions
Is n8n Assistant generally available?
No. As of September 13, 2026, n8n labels Assistant as a preview. Features, interfaces, credit use, and supported resources may change, and generated workflows may require manual correction.
Which n8n plans include Assistant?
The current documentation lists Cloud Starter and Pro plus self-hosted Community, Registered Community, and Business. Cloud and self-hosted Enterprise are not generally ready; enterprise customers can ask n8n about preview access.
Does n8n Assistant see credential secrets?
n8n says credential secrets are never sent to the AI. Users choose or create credentials through n8n's standard credential screen instead of pasting secrets into chat.
Can it fix a broken workflow?
It can inspect selected execution errors, propose or apply changes, and rerun the workflow. A successful rerun still needs human review for logic, security, unusual inputs, and side effects.
How can I reduce AI credit use?
Give a specific initial prompt, include constraints and failure behavior, review the proposed plan, start a new conversation for unrelated work, and avoid regenerating the same workflow without adding useful guidance.
Conclusion
n8n Assistant lowers the effort needed to turn an automation idea into a visible, testable workflow. Its best use is as a collaborative builder and debugger: provide a precise brief, inspect what it changes, test failure paths, and retain approval over activation and sensitive actions. Preview status means the final responsibility for correctness, security, and maintenance still belongs to the team running the workflow.