AI workflow builder for the n8n you run
Deploy your Production workflows on autopilot
Describe the automation. Get it running. We build it in your own n8n, run it for real, and fix what breaks. You just paste the credentials.
Connects in two minutes. Any n8n 2.x you host. Nothing to migrate.
When a Stripe invoice is paid, add a row to Sheets and post to Slack.
Reading the request
REV 01 / WORKFLOW BUILDER / 2026
Your workflows land in the tools you already pay for
Ten seconds to decide. Four hours to build.
Paid invoice, Sheets row, Slack ping. You knew that instantly. The wiring is what costs you the day.
The blank canvas eats the afternoon
Node search. Docs tab. One expression that keeps returning undefined.
Generic AI hands you JSON that will not import
It guesses node types your instance does not have. The import throws.
The workflow dies and nobody tells you
A token expires. The data stops. You hear about it a week later.
How it works
Three steps. None of them is wiring a node.
The workflows land in the n8n you already have. Nothing to migrate or relearn.
- 01
Paste your n8n URL and key
Two minutes. Nothing on your instance changes.
- 02
Say what you want automated
Plain sentences. Built from the nodes your instance actually has.
- 03
Walk away with it running
It runs for real, repairs what fails, and waits for you before anything live.
MCP server
Or never leave your editor at all.
Same builder, same instance, same live test. Every approval gate still applies when the caller is a model.
- Claude
- Claude Code
- Cursor
- Windsurf
Read only
- search_nodes
- get_node
- get_workflow_patterns
- get_best_practices
- list_workflows
- get_workflow
- n8n_validate_workflow
Writes, labelled as such
- n8n_create_workflow
- n8n_update_workflow
- adopt_n8n_workflow
Touches production, flagged destructive
- n8n_execute_and_diagnose
- n8n_heal_workflow
- n8n_watch_workflow
- update_native_workflow
- restore_workflow_snapshot
Version control
Ship fast without ever losing what already works.
Every change snapshots the workflow first. Break something at 4pm, be back on the working version before your coffee goes cold.
See exactly what changed before it ships
Which nodes moved, which parameters changed, what validation caught. You approve a diff you can read.
- Enrich companyHTTP Request
- Notify salesSlack
- Qualified?Threshold 60 to 70
- Workflow settingsTimeout set to 15s
“What was running last Tuesday?” has an answer
Every build writes a version and a deployment record, and neither is ever edited in place. History cannot be quietly rewritten.
One click back to working
A rollback writes a new version rather than erasing one. You recover in seconds and the trail stays honest.
Your team never overwrites each other
One shared, revision-checked draft across the app, the chat, and the MCP tools. A stale write is rejected, not silently accepted.
- 1New lead12ms
- 2Extract details19ms
- 3Enrich company842ms
- 4Qualified?4ms
- 5Notify sales216ms
Monitoring
Know it broke before your team does.
Automations die quietly. You get the broken node, its data, and the reason, usually before anyone downstream notices.
- You find out before your users do
- Every execution is watched. You get told which node moved, while there is still time.
- The reason is on screen, not buried in JSON
- You see the data the node actually received. One glance, not a midnight dump.
- Secrets never reach our database
- Credentials and payloads are redacted before anything is stored.
Guard
Broken at 3am. Fixed by the time you wake up.
Self healing sounds great until a model edits production on a hunch. Guard works to a fixed contract instead, and every boundary below is enforced in code.
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 will not turn on until all four are true
- A callable test trigger and saved, redacted test input
- Your explicit acknowledgement that repairs are replayed live
- A baseline execution that passes before anything changes
- A probe confirming your n8n exposes the APIs Guard needs
Before and after
Five jobs you do today. Here is what they cost tomorrow.
Nothing here replaces n8n. It replaces the hours you spend wiring it.
Getting one built
On n8n alone: An afternoon of node search, docs tabs, and expressions that return nothing.
With PulserNode: One sentence. Nodes that exist on your instance, wired and ready.
Knowing it works
On n8n alone: Trigger it by hand and squint at the execution log.
With PulserNode: A real execution passes before anyone calls the job done.
Shipping a change
On n8n alone: Export JSON, import it, reconnect credentials, repeat.
With PulserNode: Every build leaves a version and a deployment record you can point at.
Undoing a change
On n8n alone: Hunt for the export you hope somebody took.
With PulserNode: One click back to any version, recorded as a new version.
When it breaks later
On n8n alone: Someone reports it, or you spot the gap in the data days later.
With PulserNode: Executions are watched as they run, and repairs stay inside your limits.
Questions
The questions people ask before they connect an instance.
If yours is not here, email us and we will answer it in one reply.
What does PulserNode actually do?
You describe an automation in plain language and PulserNode builds it into the n8n instance you already run, then triggers a real execution and repairs what failed until it works. You approve the plan first, and anything with a live side effect waits for you.
How is this different from an AI that writes n8n JSON?
A general model writes node types from memory, so the JSON often fails to import. PulserNode reads the node catalog off your own instance first, builds against what is installed there, and proves the workflow by running it. You get a workflow that ran, not a file to debug.
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.
Does PulserNode replace n8n?
No. You keep your own n8n and keep editing in its editor. PulserNode connects over the public API and adds the layer around it: an AI builder, version history, monitoring, and bounded repair. There is no second editor to learn and nothing to migrate.
Do I have to give PulserNode my n8n API key?
Yes, and it is the only credential we hold. It is envelope-encrypted at rest, decrypted only for an outbound call to your instance, and never returned to the browser. Disconnect the instance at any time and our access is revoked immediately.
Which versions of n8n work with PulserNode?
Self-hosted n8n 2.x reachable over HTTPS with the public API enabled. On connect, PulserNode probes the instance and tells you which capabilities it found, rather than failing later on a call your version does not support.
Can the AI change my production workflows without asking?
No. Model output is parsed into an allowlisted set of patch operations and anything else is rejected before it reaches your instance. Changes need your explicit approval, and automatic repair only runs on workflows you enrolled yourself.
What happens if a repair makes things worse?
The patched workflow is replayed against your saved test input and compared to the baseline. A regression restores the previous snapshot. Repair stops after three attempts or fifteen minutes and hands control back to a person.
Does PulserNode store my workflow data?
It stores workflow structure, version snapshots, and sanitized execution summaries. Business payloads are not retained, diagnostics are redacted before storage, and encrypted diagnostic samples expire after 30 days.
Can I use PulserNode from Claude or Cursor?
Yes. PulserNode runs a Streamable HTTP MCP server that Claude, Claude Code, Cursor, and Windsurf all connect to. The tools call the same API the web app does, so every approval gate applies when the caller is a model.
What does it cost?
Basic is $9 a month, or $7.20 a month billed annually, and covers version history, validation, rollback, and MCP access. AI Pro is $39 a month, or $31.20 a month billed annually, and adds unlimited AI workflow building, monitoring, and automatic repair. Founding seats take 40% off while they last.