36 operations.
Across six resources: Account, Backend, Circuit, Job, Session and Workload.
Verified n8n community node
IBM Quantum, inside n8n.
Build a circuit. Run it on a real QPU. Act on the result.
Build, run and retrieve quantum circuits on the IBM Quantum Platform, straight from n8n. Verified by n8n, so it installs on n8n Cloud as well as self-hosted.
On n8n’s community nodes screen, enter
n8n-nodes-ibm-quantum
n8n Cloud or self-hosted · Free, MIT License · Unofficial, not affiliated with IBM
da45jem1vhnc73fkj6t0: 2,066 shots read 00 and 1,950 read 11, the two outcomes a Bell state should give. The other 80 are device error.Zero runtime dependencies. No Qiskit, no quantum library, nothing to compile. Circuits travel as OpenQASM 3 text, and your API key becomes a short-lived token that n8n refreshes on its own.
Get the highlights.
Three entries.
In the n8n picker: an action, a trigger for finished jobs, and an AI Agent tool that n8n builds from the action.
Hardware jobs.
Run through the node on three IBM Heron devices, every one listed in the testing record.
Tests.
With 100% coverage of the node’s logic, enforced by CI rather than reported.
Dependencies.
At runtime. No Qiskit, no quantum library, nothing to compile.
Operations
Six resources. 36 operations.
Pick a resource to see what it does, in the words n8n shows. Backend fields are dropdowns filled from your own account, labeled like ibm_kingston (Heron r2, online, 290 queued).
Get Account Configuration
Per plan on the account this instance belongs to: usage limits, allowed backends, session TTLs and functions
Get API Versions
List every IBM-API-Version and which one is still live
Get Configuration
Read the cost limit set on this instance
Get Instance
Read the plan, limits and backends attached to this instance
Get Many Instances
Every IBM Quantum instance the API key can see, with the cost limit and allowed backends of each, read from the IBM Cloud Resource Controller
Get Usage
Read seconds consumed against the quota for the current period
Get Usage Analytics
Totals for jobs, sessions and usage over a date range, with every usage and queue-time figure in milliseconds
Get Usage Analytics Filters
The backend, plan and instance values the analytics filters accept
Get Usage Analytics Grouped
The same totals, in milliseconds, split by backend, instance, plan, subscription or user
Get Usage Analytics Grouped by Date
The same totals as a time series, one point per day, in milliseconds
Set Cost Limit
Cap the QPU seconds this instance may spend, or clear the cap, through the IBM Cloud Resource Controller
Get Configuration
Static device data: basis gates, coupling map and qubit count
Get Defaults
Pulse level defaults; every IBM Cloud device returns an empty object
Get Least Busy
Pick the online device with the shortest queue
Get Many
Every device on this instance, with status, qubits and queue
Get Properties
Calibration data: gate errors, readout errors and timings
Get Status
Live availability, queue depth and any operator message
Build
Assemble an OpenQASM 3 program from a gate list, validated locally
Import OpenQASM 3
Pass an existing OpenQASM 3 program through, header checked
Cancel
Stop a queued or running job; queued jobs cost nothing
Delete
Remove a job and its results permanently
Get Logs
Read whatever the program wrote to the job log
Get Many
Recent jobs, newest first, without their circuit payloads
Get Many Tags
Search the tags in use; the term must be 3 to 100 characters
Get Metrics
Timestamps, QPU seconds charged and execution time
Get Results
Wait for a terminal state, then parse counts or expectation values
Get Status
The current state of a job, with the failure reason when it has one
Submit to Estimator
Run a circuit and read expectation values of observables
Submit to Noise Learner
Characterize the noise of a circuit layer
Submit to Sampler
Run a circuit and read measurement counts
Update Tags
Replace the whole tag list on a job
Close
End the session at once: it stops accepting new jobs, queued jobs will not run, and a running job finishes. To let queued jobs drain first, use Set Accepting Jobs instead.
Create
Open a batch or dedicated session on one backend
Get
Read a session state and its remaining time
Set Accepting Jobs
Stop a session taking new jobs; that ends it and it cannot reopen
Get Many
Jobs and sessions in one listing, with richer filters
Real hardware
A real chip runs only its own gates.
The Qiskit Runtime REST API does not transpile. A textbook h or cx is queued, then fails with reason code 1517, and a failed job still spends quota.
Runs as-is on Heron 10
xsxrxrzrzzczmeasureresetdelaybarrier
Transpile first 16
yzhssdgttdgrypucxswapcrxcrycrzccx
Accepted, writes nothing 1
id
IBM lists id as native, yet a job carrying it fails, so the node leaves it out of the program.
The 27 entries of the Circuit › Build palette. Every OpenQASM 3 Submit reads the chosen backend’s basis gates and coupling map first, and warns about an instruction or a qubit pair the device cannot run before the job is queued.
Build it native.
Two identities turn a Bell state into gates the chip runs, with no transpiler at all:
H = rz(π/2) · sx · rz(π/2)CNOT = H · cz · H
Measured on ibm_fez over 2,048 shots: 48.7% 00 and 46.2% 11.
OPENQASM 3.0;
include "stdgates.inc";
qubit[2] q;
bit[2] c;
rz(1.5707963267948966) q[0];
sx q[0];
rz(1.5707963267948966) q[0];
rz(1.5707963267948966) q[1];
sx q[1];
rz(1.5707963267948966) q[1];
cz q[0], q[1];
rz(1.5707963267948966) q[1];
sx q[1];
rz(1.5707963267948966) q[1];
c[0] = measure q[0];
c[1] = measure q[1];Or transpile it, free.
Transpile locally with Qiskit, on any plan, then paste the result into the node. A fake backend carries the real device’s topology and gates, so no credentials are needed.
Pin the Submit to the device you transpiled for.
from qiskit import QuantumCircuit, qasm3
from qiskit.transpiler.preset_passmanagers import generate_preset_pass_manager
from qiskit_ibm_runtime.fake_provider import FakeFez # mirrors ibm_fez
backend = FakeFez()
qc = QuantumCircuit(2, 2)
qc.h(0)
qc.cx(0, 1)
qc.measure(0, 0)
qc.measure(1, 1)
isa = generate_preset_pass_manager(optimization_level=1, backend=backend).run(qc)
print(qasm3.dumps(isa)) # paste this into the nodeLong-running jobs
Submit now. React when it’s done.
A hardware job can sit in the queue for hours, and IBM offers no push or callback, so something has to poll. The trigger polls on n8n’s scheduler, not inside a held-open execution.
Workflow one
- Build
- Submit to Sampler
Submits and finishes at once, with the jobId. Nothing blocks.
Workflow two, active
- IBM Quantum Trigger
- Get Results
Fires when a job reaches a terminal state. Get Results returns at once, because the job is already done.
Trigger On
- Any Terminal (Completed, Failed or Canceled)
- Completed
- Failed
- Canceled
- Failed or Canceled
- Failed or Canceled is for alerting: every event carries
reason,reasonCodeandreasonSolution, so a workflow can page someone or resubmit elsewhere. - Restarts are safe. The jobs already fired live in the workflow’s static data, so a restart repeats none of them, and its first poll fires the ones that finished while n8n was down, up to Jobs to Scan.
- Tags keep workflows apart. Set them on Submit and filter on them in the trigger, and each workflow reacts only to its own jobs.
AI Agent tool
Ask it in plain words.
Attach the node to an n8n AI Agent and the model calls its operations directly. It fits questions best, where the agent asks and gets a structured answer.
Which QPU has the shortest queue right now?
Backend › Get Least Busy
Which Heron device will start my job soonest?
Backend › Get Least Busy, Rank By an estimated wait, Processor Family Heron
How much runtime is left on my instance?
Account › Get Usage
Which backends and session limits does my plan allow?
Account › Get Account Configuration
Did job d1abc finish?
Job › Get Status, with Include Circuit Params off
Submission works too, but on hardware the agent has to supply a transpiled circuit, so pair it with a circuit built beforehand rather than asking the model to write one.
Quota
QPU seconds are the budget.
The Open plan gives 600 seconds of QPU time per rolling 28 days, and a failed job still spends some of it. The node gives you the brakes.
Max Cost
Caps what one job may spend. Left at zero, IBM stamps the job with the plan maximum, on the Open plan all 600 seconds, so set a real number.
Set Cost Limit
A ceiling for every job on the instance, written through IBM Cloud’s Resource Controller and read back from it. Clear Limit returns the instance to the plan default, on the Open plan 600 seconds.
Sessions and batches
Keep the jobs of a hybrid loop together instead of sending each back to the general queue. Batch is the only mode the Open plan allows.
Usage
Get Usage compares seconds consumed with the limit; the analytics operations break the spend down by backend, plan, user or date.
In the editor
What it looks like in n8n.
Tested
Run on real hardware, and written down.
Every job this project has run on IBM hardware through the node is in its testing record, failed and canceled ones included: most of those were sent to measure exactly how IBM refuses something.
283 jobs
187 completed, 68 failed and 28 canceled, between June 24 and September 10, 2026.
1,799 QPU seconds
30.0 minutes of real device time, as IBM charged it.
3 Heron r2 devices
ibm_fez, ibm_marrakesh and ibm_kingston, 156 qubits each.
1,161 unit tests
In 25 files, holding the node’s logic at 100% of statements, branches, functions and lines, a gate CI enforces.
Security
Four hosts, and no others.
n8n keeps your API key as an encrypted credential. The node reads it in one place, to exchange it with IBM for a short-lived token, and every API call it makes after that has a 30-second timeout.
| Host | Used for |
|---|---|
iam. | Exchanging the API key for a short-lived token |
quantum. | Qiskit Runtime, US East instances |
eu-de. | Qiskit Runtime, EU (Germany) instances |
resource-controller. | IBM Cloud Resource Controller, only for Get Many Instances and Set Cost Limit |
Watch and read
From setup to the first result.
My IBM Quantum node for n8n is now live
The launch, the n8n verification, and what the three nodes do.
Running Quantum Circuits on Real IBM Hardware from n8n
An end-to-end Bell state on a real QPU, including the transpilation step that trips up most first attempts.
Six defects in a verified n8n node for IBM Quantum
What 121 seconds of real QPU time found that green tests and 98 percent coverage could not.
One circuit, two IBM Quantum error codes: 1506 and 1603
The same rejected program, submitted twenty times, came back as one code eleven times and as the other nine.
Questions.
What is n8n-nodes-ibm-quantum?
An unofficial, community-maintained n8n node for the IBM Quantum Platform. It builds OpenQASM 3 circuits, submits them to Qiskit Runtime on real IBM hardware, reads the results, and starts workflows when jobs finish. n8n has verified it.
Is it made by IBM?
No. It is unofficial and community-maintained, and it is not affiliated with, endorsed by, or sponsored by IBM. Every node carries an (Unofficial) marker in the n8n picker.
Does it work on n8n Cloud?
Yes. It is verified by n8n, so it installs on n8n Cloud as well as on self-hosted n8n: open the community nodes screen and enter the package name, n8n-nodes-ibm-quantum.
What do I need to use it?
An IBM Cloud account with access to the IBM Quantum Platform, an IBM Cloud API key, and the Cloud Resource Name (CRN) of your Qiskit Runtime instance. Self-hosted n8n needs Node.js 24 from n8n 2.36 on; Node.js 22 is enough for n8n 2.35 and older.
Do I need Qiskit?
Not to run the node: it has zero runtime dependencies. Real hardware accepts only transpiled circuits, though, so for anything beyond the native gates you transpile locally with Qiskit, free on any plan, and pass the result in.
Why did my job fail with reason code 1517?
The circuit was not transpiled to the backend’s native gates, or a two-qubit gate landed on a pair the chip does not connect. Build it from native gates, or transpile it for the exact device you submit to. The node warns about both before the job is queued.
Is there a simulator?
No. IBM retired its cloud simulators on May 15, 2024, so every backend is real hardware. To try a circuit without spending quota, simulate it locally with Qiskit first.
Can an AI Agent use it?
Yes. The action node is usable as an n8n AI Agent tool, and on n8n 1.85.0 or newer nothing extra is needed. It fits questions best, such as which QPU has the shortest queue or how much runtime is left.
Is it free?
The node is free and open source under the MIT License. Jobs use your IBM Quantum plan: the Open plan gives 600 seconds of QPU time per rolling 28 days, and a failed job still spends some of it.
Put IBM Quantum in a workflow.
Free and open source, under the MIT License.
n8n-nodes-ibm-quantum