Recipe 3: connect Jev to n8n
Start with a generic HTTP Request node and an explicit review branch.
Begin without production integrations
Use a Manual Trigger and fixed sample text before connecting a live Webhook.
Manual Trigger → Edit Fields (test email) → HTTP Request → Switch / If → review record.
This configuration follows the documented HTTP interface. We have not logged into your n8n instance or executed a live workflow. Node names may differ by version.
Configure HTTP Request
| Setting | Value |
|---|---|
| Method | POST |
| URL | https://api.typesafe.ai/v1/systemone |
| Authentication | Store Header Auth in n8n credentials |
| Credential header | Authorization: Bearer followed by your TypeSafe key |
| Body content type | JSON |
| Response | JSON, with an explicit error path |
{
"model": "jev-1.13.0",
"state": "My parcel has not arrived. Please check delivery.",
"questions": {
"queue": {
"type": "choice",
"instructions": "Choose the support queue. Use other when unclear.",
"criteria": {
"shipping": "Delivery status or missing parcel",
"refund": "Explicit refund request",
"billing": "Payment or invoice issue",
"other": "Unclear or another topic"
}
}
}
}
Route the response
When the node returns only the response body, inspect answers.queue.choice and answers.queue.confidence. Enabling the full-response option may add a body wrapper. Check the execution preview rather than copying a field path blindly.
Map allowed labels to queues. Errors, missing fields, unknown labels and results below a validated threshold must go to review. The first workflow should create records, not send replies.
Before connecting real records
Replace fixed state with authorized, redacted and length-limited input from the previous node. Keep the API key in credentials, not shared JSON, prompts, frontend pages or exported samples.
Sources and verification boundary
Reviewed September 19, 2026. Documentation-based guidance, not a live API or business-performance test.
n8n · HTTP Request node →TypeSafe · HTTP API →TypeSafe · Confidence →
