How it works
From one sentence to a workflow that ran.
Four stages, all of them in the n8n you already run. You approve the plan, anything with a live side effect waits for you, and every change is reversible.
- 1.0 Connect
Paste your n8n URL and an API key.
PulserNode tests the connection before saving it and tells you what it found: the n8n version, whether the public API answers, and which capabilities are available. Nothing on your instance changes.
- The key is envelope-encrypted and never returned to a browser.
- Disconnect at any time and access is revoked immediately.
Sample · connect your instanceStep 1 of 3 - 2.0 Describe
Say what should happen, in a sentence.
Pulser AI reads the node catalog on your instance and plans the steps from nodes you actually have. If something is missing, like a Slack channel or a sheet name, it asks for exactly that value.
You approve the plan before anything is written.
Sample · the plan3 steps When a Stripe invoice is paid, add a row to Sheets and post to Slack.
- Invoice paidStripe triggerPlanned
- Add a rowGoogle SheetsPlanned
- NotifySlackPlanned
- 3.0 Build and live test
It is built in your n8n, then run for real.
The workflow is written through n8n's public API and fired with a real execution. When a node fails, the builder reads what n8n recorded, patches the one parameter at fault, and runs it again, up to three times.
A workflow without a testable trigger finishes honestly as untested, never as passed.
Sample · live testPassed - Invoice paidStripe triggerPassed14 ms
- Add a rowGoogle SheetsPassed612 ms
- NotifySlackPassed208 ms
- Invoice paid
stripeTriggerPassed14 ms1 item - Add a row
googleSheetsPassed612 ms1 item - Notify
slackPassed208 ms1 item
- 4.0 Watch and heal
When it breaks later, Guard handles the first response.
Executions are watched as they run. A failure is diagnosed from the data the node received, patched with an allowlisted operation, replayed against your saved input, and kept only if it passes. A regression restores the version that worked.
- Diagnose. Read the runtime data the failing node actually received. No guessing.
- Patch. Allowlisted operations only. A model cannot invent a node type or write raw code.
- Replay. Run the fix against the test input you saved. Nothing ships on an argument.
- Verify. A pass keeps the fix. A regression restores the version that worked.
Guard turns on only when all four are true
- 01A callable test trigger and saved, redacted test input exist
- 02You have acknowledged that repairs are replayed live
- 03A baseline execution passes before anything changes
- 04A probe confirms your n8n exposes the APIs Guard needs
Credentials and payloads are redacted before anything is stored.
Questions about the walkthrough.
Not here? Email hello@pulsernode.com
What kinds of automations can it build?
Anything the nodes on your instance support. There is no fixed template library: PulserNode searches your own node catalog for each request, so its range grows with the integrations you install. Anything with a live side effect pauses for your approval.
What does PulserNode need from my n8n instance?
An HTTPS address our servers can reach and an API key from n8n's own settings, with the public API enabled. The connection is tested before it is saved, and the test reports the n8n version and which capabilities it found.
What happens between typing a sentence and a running workflow?
Pulser AI plans the steps from your instance's node catalog, writes the workflow into n8n through its public API, and fires a real execution. If a node fails it patches the one parameter at fault and runs it again. Anything with a live side effect waits for your consent.
Can I stop halfway and finish in n8n myself?
Yes. The workflow exists in your instance from the moment it is written, so you can open it in the n8n editor at any point. PulserNode records a snapshot before each change, so switching between the two never loses work.
Your instance, your workflows, one sentence at a time.
Connect your n8n, describe the first one, and get back to your actual work.