What you build
The canvas is a visual builder. You drag nodes from a palette of around seventy-five keyword types across roughly fourteen categories, connect them into a sequence, configure each one, save the workflow, and run it. A run drives a real pooled browser and produces per-node results, a run history and a downloadable trace, so a passing run reflects work that actually happened.
RPA: capture, fill, verify, decide
- Capture Page records the current page as a structured record, including the values in the fields, not just the markup.
- Fill From Record types a stored record back into a form.
- Verify Record compares what is on the page against a record and reports the differences.
- Decision Table routes the workflow down a branch based on the data, using a table you can read rather than a tangle of conditions.
Connectors use your own credentials
Connector nodes for Slack, Telegram, WordPress, Reddit, Google Sheets and web search call the real third-party service using your own credentials, stored in the credential vault. Each one verifies the account first and returns a clear error rather than a silent empty result. A node that writes to an external system is treated as a high-risk step and is never run on its own by an automated planner. Outbound traffic is restricted to allow-listed hosts.
The local AI prompt node
An AI prompt node runs on a local model on the platform through a governed gate, and makes no external AI call. It is a generation, labelled as unverified, not a source of truth, and under load it can be refused by the concurrency cap. Use it to draft or classify inside a workflow, not to assert a verdict.
Run it, schedule it, import into it
Run a workflow on demand, on a schedule, or from a secret-gated webhook, and read the results and traces in the results tab. You can also import an exported n8n workflow as a starting point: unsupported node types are reported and embedded secrets are stripped, capped at sixty nodes. A workflow you do not own returns not-found, so nothing leaks across tenants.
Questions people ask about this
- Does the AI node send my data to an external model?
- No. It runs on a local model on the platform, behind a governed gate, and makes no external AI call. Its output is labelled as generated and unverified.
- Can the quiz node answer a quiz for me?
- No. The quiz node checks whether a quiz application actually works, that every question is answerable and grading is correct. It is a test of the app, not an exam-answering bot.
Source: /documentation/workflow-canvas, updated 2026-09-18. Answers are checked against the shipped product on that date.
Found something out of date? The people who build the module answer the contact form on the home page, and corrections to this guide are dated above. Tell us.