Integration guide · n8n · published October 11, 2026
Jev in n8n
Classify, score and route every item in an n8n workflow with Jev, and send only the unsure ones to a person.
n8n workflows are full of branching decisions: which team gets this ticket, is this lead worth a call, does this message need a reply today. Jev answers each one with a probability in a fraction of a second, so the workflow can route the sure cases and queue the rest. There is no official node yet, but 7 community nodes cover it, with about 13,865 npm downloads between them last month. We compared them from their published code and ran one end to end.
Set it up in four steps
The same steps apply to every node below; names of fields vary slightly.
- 01
Install the community node
In n8n, open Settings → Community Nodes → Install and enter the package name, for example n8n-nodes-jev-classification. Unverified community packages install on self-hosted n8n only.
- 02
Create the credential
Add the node’s credential with your API key. For a Jev AI key, set Base URL to the Jev AI address in the table below; for a TypeSafe key, leave the default. The credential test calls the models endpoint, so a wrong address fails here, not mid-workflow.
- 03
Describe the decision
Pick Classify, Score, Check or Ask. Give each category a description of when it applies; the description is what Jev reads, so write it the way you would brief a new teammate.
- 04
Set the threshold and branches
Each category becomes an output, plus Needs Review for anything below your confidence threshold. Wire Needs Review to a person, a Slack message or a slower LLM step.
The Jev community nodes, compared
Monthly npm downloads and the Base URL each credential expects, as of October 11, 2026.
| Package | Downloads / month | What it does | Default address | Base URL for a Jev AI key |
|---|---|---|---|---|
| @taifoon/n8n-nodes-typesafev2.0.5 · node “Taifoon TypeSafe” | 7,408 | Ask yes/no, pick-one and rate-it questions about any item and route on the answers; a Translate operation compiles a plain-language task into questions offline. | https://api.typesafe.ai (credential “Base URL”, node appends /v1/systemone) | https://jev-ai.pro/api |
| n8n-nodes-typesafev0.3.3 · node “TypeSafe” | 2,152 | Typed questions with an output mode of one branch per question or one item carrying every answer plus token usage. | https://api.typesafe.ai/v1 (node appends /systemone) | https://jev-ai.pro/api/v1 |
| n8n-nodes-jev-aiv1.0.8 · node “Jev AI” | 1,553 | Multi-branch routing and policy guardrails, also usable as a tool by n8n’s AI Agent. | https://api.typesafe.ai/v1 | https://jev-ai.pro/api/v1 |
| n8n-nodes-jevv0.2.3 · node “Jev” | 1,127 | Classify, route and score text with questions you define, with a probability for every answer. | https://api.typesafe.ai (credential test calls /v1/models) | https://jev-ai.pro/api |
| n8n-nodes-jev-classificationv0.3.0 · node “Jev Classification” | 855 | Classify, Score, Check and Ask operations; one output per category plus a Needs Review output below a confidence threshold; parallel requests and several items per request. | https://api.typesafe.ai (node appends /v1/systemone; placeholder suggests OpenRouter) | https://jev-ai.pro/api |
| n8n-nodes-jev-routerv0.1.2 · node “Jev Router” | 548 | Classify, score and route workflow items with typed decisions. | https://api.typesafe.ai | https://jev-ai.pro/api |
| n8n-nodes-jev-gatev0.1.0 · node “Jev Gate” | 222 | A confidence gate: act on sure answers, send the rest to review, and calibrate the threshold on your own data. Defaults to OpenRouter as provider. | Provider setting (OpenRouter or TypeSafe) plus an optional Base URL | Provider TypeSafe, then https://jev-ai.pro/api |
All seven are marked usable as tools, so n8n’s AI Agent node can call them too. Not in the table: n8n-nodes-jev-dual-engine (2,432 downloads). Despite the name, this is a chat-model node for OpenAI-compatible and Gemini APIs (its credential defaults to a local Ollama /v1 chat endpoint). It does not call Jev’s /v1/systemone decision API.
Our run: 40 support messages through Jev Classification
Real n8n, a public labelled dataset, nothing tuned.
| n8n | 2.42.6, self-hosted on a laptop, workflow executed with n8n execute |
|---|---|
| Node | [email protected], Classify, threshold 0.8, Needs Review on, 4 parallel requests |
| Model | jev-1.13.0 through TypeSafe’s API, 2026-10-11 |
| Data | Banking77 test split: 5 intents × 8 messages, every 5th message of the 40 per intent |
| Categories | card_arrival Card arrival: a new or replacement card has not arrived yet, or asking when or where it islost_or_stolen_card Lost or stolen card: the card was lost, stolen or taken and needs blockingrequest_refund Refund request: the customer wants money back for a purchase or chargetransfer_not_received_by_recipient Transfer not received: money was sent but the recipient says it has not arrivedexchange_rate Exchange rate: questions about currency exchange rates or the rate applied |
| Output branch | Items |
|---|---|
| card_arrival | 7 |
| lost_or_stolen_card | 8 |
| request_refund | 8 |
| transfer_not_received_by_recipient | 8 |
| exchange_rate | 8 |
| Needs Review | 1 |
- 39 of 39 routed messages (100.0%) landed in the right branch. Every one of them had a top probability of at least 0.8.
- The one in Needs Review deserved to be there. “How do I locate my card?” is labelled
card_arrivalin the dataset; Jev split it lost_or_stolen_card 0.58 and card_arrival 0.42. A person would ask a follow-up question too. - 10.0 seconds for 40 items with 4 parallel requests, the node’s default.
- Limits: 40 messages and 5 well-separated intents is an easy, small test. Close intents and messier text will push more items into review, which is the point of the branch. Our Banking77 comparison covers all 77 intents.
No community node: the HTTP Request node
Works on n8n Cloud and self-hosted, with nothing to install.
{
"parameters": {
"method": "POST",
"url": "https://jev-ai.pro/api/v1/systemone",
"authentication": "genericCredentialType",
"genericAuthType": "httpHeaderAuth",
"sendBody": true,
"specifyBody": "json",
"jsonBody": "={{ JSON.stringify({ model: 'jev-latest', state: { message: $json.text }, questions: { intent: { type: 'choice', instructions: 'Which team should handle this message?', criteria: { billing: 'Charges and refunds', cards: 'Lost, stolen or missing cards', transfers: 'Payments that did not arrive' } }, urgent: { type: 'noul', instructions: 'Does the customer need help today?' } } }) }}"
},
"name": "Ask Jev",
"type": "n8n-nodes-base.httpRequest",
"typeVersion": 4.2
}Create a Header Auth credential with name Authorization and value Bearer followed by your Jev AI key, and select it on the node. The response has answers.intent.choice, a probability per option and answers.urgent.noul. Put an If or Switch node after it with a condition such as:
{{ $json.answers.intent.probabilities[$json.answers.intent.choice] >= 0.8 }}You lose the ready-made branches and batching of the community nodes, and you gain full control of the request: any number of questions, any question type, and the same body you tested in the playground.
Workflows people build with it
- Support triage: classify the inbound email, score urgency, open the ticket in the right queue. See support ticket triage.
- Lead routing: score form submissions against your ideal-customer description, send hot leads to the CRM and sales Slack. See lead qualification.
- Content gates: a yes/no check for spam, abuse or prompt injection before an LLM step reads the text. See prompt injection detector.
- LLM routing: decide whether a request needs a large model or a small one before the expensive call. See LLM router.
For agent frameworks, see Jev in an agent loop and Jev in OpenClaw; for more open-source builds, Jev examples.
Jev and n8n FAQ
Is there a Jev node for n8n?
Not an official one from n8n or TypeSafe, but several community nodes, installable on self-hosted n8n. The most downloaded on October 11, 2026 were @taifoon/n8n-nodes-typesafe, n8n-nodes-typesafe, n8n-nodes-jev-ai, n8n-nodes-jev and n8n-nodes-jev-classification. You can also call Jev from n8n’s built-in HTTP Request node with no install.
Which n8n Jev node should I use?
For routing items into branches with a review queue, Jev Classification: one output per category, a Needs Review output below a threshold, parallel requests and batching. For several questions per item with one branch per answer, n8n-nodes-typesafe. For the most widely installed general node, @taifoon/n8n-nodes-typesafe. All of them can also act as tools for n8n’s AI Agent.
Can I use a Jev AI API key in n8n?
Yes. Every node in the table has a Base URL field on its credential. Set it to https://jev-ai.pro/api for nodes that append /v1/systemone themselves, or https://jev-ai.pro/api/v1 for nodes whose default ends in /v1, and paste your Jev AI key. The credential test confirms the address before any workflow runs.
How accurate is Jev in an n8n workflow?
In our run on 40 Banking77 support messages across 5 intents, the Jev Classification node routed 39 with confidence above 0.8 and all 39 were correct; the one remaining message, “How do I locate my card?”, went to Needs Review. Forty messages is a small sample, so measure on your own data before trusting a threshold.
How fast is it?
The node classified 40 items in 10.0 seconds with 4 parallel requests against hosted Jev from Asia, about a quarter of a second of wall time per item. The nodes that batch several items per request can go faster on large runs.
Does n8n-nodes-jev-dual-engine call Jev?
Despite the name, this is a chat-model node for OpenAI-compatible and Gemini APIs (its credential defaults to a local Ollama /v1 chat endpoint). It does not call Jev’s /v1/systemone decision API.
Can I use Jev with n8n Cloud?
n8n Cloud allows only community nodes that n8n has verified. If the Jev node you want is not on that list, use the HTTP Request node with a Header Auth credential, which works on Cloud and self-hosted alike.
About this page
Who. Jev AI runs a hosted endpoint for TypeSafe’s Jev and publishes guides and measurements around it. We did not write any of these nodes and are not affiliated with n8n, TypeSafe or the node authors.
How. We downloaded each package from npm and read its credential fields and request paths from the published code; downloads are npm’s last-month counts. The workflow ran on a fresh self-hosted n8n with the credential imported through n8n’s CLI, so the key was encrypted by n8n and never written into the workflow.
When. Published October 11, 2026. Community nodes update often; check the version on npm before installing.
Sources
n8n is a trademark of n8n GmbH. Community nodes belong to their authors.