@humangent/n8n-nodes-humangent

Official n8n node for Humangent — a human-in-
the-loop inbox for n8n workflows. Drop the Humangent node onto a
canvas, pick a task type, map fields, and the workflow pauses on a
human decision and resumes along the branch that matches the
reviewer's outcome.

Install

In your self-hosted n8n instance:

Settings → Community Nodes → Install → @humangent/n8n-nodes-humangent

Restart n8n if prompted. Minimum supported n8n version: 1.70.

Credential setup

Create a Humangent API credential in n8n. One field, required,
issued from Humangent's admin console at /admin/api-keys:

The key is shown once when you create it — copy it before closing
the modal. Use an hmk_test_* key for the Test step button so
test-run requests stay tagged separately from production audit trails.

Click Test on the credential dialog to verify.

Test mode

Use an hmk_test_* API key in the credential when running n8n's
Test step flow. Test-mode requests are still sent to Humangent but
are marked separately from live audit trails. Use an hmk_live_* key
only for production workflows.

Node walk-through

Task Type

A resource-locator with a From list (searchable dropdown) only.
Pick the task type from the list — the node captures a snapshot of
the task type's outcomes at pick time and renders one canvas branch
per outcome (plus a Dismissed lane and a Timed Out lane).

The snapshot is the canvas truth source. If the task-type author later
adds, removes, or relabels outcomes upstream, the canvas keeps showing
what you picked. To refresh, re-open the dropdown and re-pick the
task type — that re-captures the snapshot.

Fields

A resource-mapper bound to the task type's field schema. Upstream
keys that match field names are auto-mapped; override or provide
expressions per field.

Limit Wait Time + Wait Time Unit

How long n8n waits before routing to the Timed Out branch.
Default 24 hours.

Important: n8n's own EXECUTIONS_TIMEOUT_MAX env var caps the
workflow-wide wait. The default is 3600 s (1 hour). If your
Limit Wait Time exceeds that cap, n8n can end the execution before a
reviewer decides. Raise the env var on your n8n instance:

export EXECUTIONS_TIMEOUT_MAX=604800   # e.g., 7 days
# restart n8n

EXECUTIONS_TIMEOUT_MAX=0 disables the cap entirely.

Output branches

Output ordering on the canvas (N outcomes in the snapshot captured
when you picked the task type):

All non-timeout branches receive a JSON item of this shape:

{
  "delivery_id": "<opaque id>",
  "request_id": "<uuid>",
  "outcome_id": "<outcome or 'dismiss'>",
  "is_dismiss": false,
  "fields": {
    /* final field values */
  },
  "fields_before": {
    /* pre-edit values, only when reviewer edited */
  },
  "decided_by_profile_id": "<uuid>",
  "decided_at": "2026-04-23T12:34:56Z",
  "duration_ms": 12345,
  "is_test": false,
}

The Timed Out branch's item is narrower — no reviewer acted, so
fields carries the last saved state and
fields_before/decided_by_profile_id/decided_at are absent.

Handling outcome drift

The node captures a snapshot of the task type's outcomes when you pick
it from the dropdown. If the task-type author adds or renames outcomes
upstream, the canvas keeps rendering the snapshot — so a reviewer can
pick a live outcome the snapshot doesn't know about. When that
happens, the node routes the decision onto Dismissed with two
extra fields on the JSON payload:

{
  "outcome_id": "needs_changes",
  "drift_detected": true,
  "unmatched_outcome_id": "needs_changes",
  // …rest of the decision payload as usual
}

You can branch on drift_detected === true downstream of Dismissed to
escalate, log, or re-route. Decisions are never silently dropped.

To refresh the canvas to match the live outcomes, re-open the
task-type dropdown and re-pick the task type.

Detached mode (Create + Continue)

Long-running reviews — anything that risks blowing past
EXECUTIONS_TIMEOUT_MAX (defaults to 1 hour on n8n Cloud) — need
the Create mode plus a Humangent Continue trigger node in a
separate workflow. The original Create and Wait mode stays the
right choice for short reviews; detached mode exists for waits that
shouldn't be bounded by a single execution's wall time.

When to use which

Bootstrap sequence (admin → builder)

1. Org admin opens Workspace Settings → n8n Allowed Origins
   in the Humangent app and adds the origin (scheme://host[:port])
   of every n8n instance that will register Continue nodes
   (e.g., https://n8n.acme.com).
2. Builder creates the destination workflow and drops a
   Humangent Continue node into it. Pick the Task Type that
   matches what the source workflow will create. Save and activate
   the workflow — activation registers the Continue with Humangent.
3. Builder returns to the source workflow. On the Humangent
   action node, set Mode → Create, fill in Continuation
   Workflow (the destination workflow you just activated) and
   Continue Node Name (the n8n-internal name of the Continue —
   visible in the destination canvas). Save.
4. Run a test step on the source workflow. The execution hint echoes
   the resolved Continue display name + Task Type — that's your
   confirmation the destination resolved correctly. Real review
   testing happens on the destination workflow's own listen-for-event
   test step.

The source emits a single Main output of
{ requestId, requestUrl, expectedTimeoutAt } and continues
immediately. The destination Continue's per-outcome / Dismissed /
Timed Out branches fire the same payload shape inline Create and Wait emits — see Output branches above.

Security

Decision deliveries from Humangent are HMAC-SHA256 signed with your
Humangent API Key. The node verifies every delivery before routing
onto a branch; invalid signatures cause the execution to stay waiting
and Humangent's outbox to retry.

Rotate keys at any time from /admin/api-keys — issue a new one,
update the credential, revoke the old.

Long-running requests

Humangent makes no SLO commitment on how long a request stays open.
Reviewer chains can extend a request from minutes to weeks (especially
when multi-level approval kicks back to a prior level for revision).

Always drive workflow resumption from the webhook callback, not from
a Wait node with a deadline timeout. Set up your Humangent node to
send the request, then wait for the decision-delivery webhook to fire
— the webhook arrives whenever the decision is final, regardless of
how many hours/days/weeks elapsed.

If you need a hard time bound, configure auto-approve mode on the
task type (Humangent will finalize the request automatically after the
configured timeout) — don't try to enforce it via n8n's Wait node.

Example workflow

Short review:

1. Trigger node receives the business event.
2. Humangent node uses Create and Wait.
3. Pick a Task Type and map fields from the trigger payload.
4. Connect each outcome branch to the matching downstream action.
5. Connect Dismissed and Timed Out to explicit handling paths.

Long review:

1. Destination workflow starts with Humangent Continue and is
   activated after selecting the same Task Type.
2. Source workflow uses Humangent Create mode.
3. Source Continuation Workflow points at the destination workflow.
4. Source Continue Node Name matches the destination Continue node.
5. Source workflow continues immediately; the destination workflow runs
   when Humangent delivers the decision.

Troubleshooting

Support

Docs and contact: humangent.io.