Calling Jev Through the Gateway - Barndoor Documentation

Jev is TypeSafe’s System One model. It is a classifier, not a chat model: you send it some content (the state) and a set of typed questions, and it answers each question with probabilities. For example, “is this ticket about billing?” comes back as 0.97, not as a sentence. The gateway serves Jev on TypeSafe’s own endpoint, POST /v1/systemone, with TypeSafe’s request and response format unchanged. So TypeSafe’s Python and JavaScript SDKs work against Barndoor once you change their base URL and API key.

Jev answers only on /v1/systemone. It does not answer /chat/completions, /responses, or /messages. On My Models, Jev routes and models carry a Classifier badge so they aren’t mistaken for chat models.

Before you start

An admin needs to make Jev available in LLM Management. There are two ways to serve it:

TypeSafe AI provider. Add the TypeSafe AI provider with a TypeSafe API key from the TypeSafe console, and enable jev-latest, or a pinned version such as jev-1.13.0 if you need answers that don’t change when TypeSafe releases a new model.
OpenRouter. OpenRouter also serves Jev, as ~typesafe/jev-latest. Enable that model on an OpenRouter provider.

Optionally, create a model route for Jev (for example jev) so callers use one stable name, and the route can fail over between TypeSafe and OpenRouter. A few rules apply, and the gateway enforces them when you save:

A route holds Jev or chat models, never both. A request that failed over from one kind to the other would fail, so the gateway refuses a mixed route. In the Create Model Route and Add Targets dialogs, models of the other kind are disabled.
Jev can’t be a routing-policy slot. A slot has to answer chat.
Jev can be a routing policy’s determiner. See Routing Policies.

Find the name to call

Open Settings → My Models. A Jev route, or a Jev model listed under Standalone Models, carries a Classifier badge. Use its name in the model field: the route name (for example jev), or <provider name>/<model> for a standalone model. The copy button next to each name gives you the exact string. When you can see a Jev model, Gateway Endpoint also lists the System One URL:

https://app.barndoor.ai/api/llm-gateway/v1/systemone

Call Jev

Authenticate with a Barndoor gateway API key (bd-…), the same as for chat models. TypeSafe’s SDKs add /v1/systemone themselves, so give them the gateway URL without /v1: https://app.barndoor.ai/api/llm-gateway.

Python
JavaScript
cURL

Install TypeSafe’s SDK with pip install typesafe-sdk (or uv add typesafe-sdk).

import os

from typesafe_sdk import Noul, TypeSafeClient

with TypeSafeClient(
    api_key=os.environ["BARNDOOR_API_KEY"],
    base_url="https://app.barndoor.ai/api/llm-gateway",
    model="jev",  # your Jev route, or "<provider name>/jev-latest"
) as client:
    result = client.system_one(
        "I was charged twice. Please fix this ASAP.",
        {"billing": Noul(instructions="Is this ticket about billing?")},
    )
    print(result.nouls["billing"].noul)

You can also set TYPESAFE_BASE_URL and TYPESAFE_API_KEY in the environment instead of passing them.

Install TypeSafe’s SDK with npm install @typesafe-ai/sdk (Node.js 20 or newer).

import { choice, TypeSafeClient } from "@typesafe-ai/sdk";

const client = new TypeSafeClient({
  apiKey: process.env.BARNDOOR_API_KEY,
  baseURL: "https://app.barndoor.ai/api/llm-gateway",
  defaultModel: "jev", // your Jev route, or "<provider name>/jev-latest"
});

const response = await client.systemOne({
  state: { document: "I was charged twice. Please fix this ASAP." },
  questions: {
    category: choice("What is this ticket about?", {
      billing: null,
      technical: null,
      other: null,
    }),
  },
});

console.log(response.answers.category.choice);

curl https://app.barndoor.ai/api/llm-gateway/v1/systemone \
  -H "Authorization: Bearer $BARNDOOR_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "jev",
    "state": "Help! My payouts have been failing for 3 days.",
    "questions": {
      "is_urgent": { "type": "noul", "instructions": "Does this convey urgency?" }
    }
  }'

Replace app.barndoor.ai with your own Barndoor host if you have one. The request body and the answers are TypeSafe’s format: see TypeSafe’s API reference for the question types (noul, choice, score) and how to read the answers.

Usage and cost

Jev requests appear in LLM usage reporting like any other model, with the same user, API key, and budget attribution. TypeSafe charges for input tokens only, so Jev’s output tokens are recorded but priced at zero.

Troubleshooting

A Jev model fails when called from a chat client

Jev answers only on /v1/systemone. Sending its name to /chat/completions, /responses, or /messages returns a 400 saying the provider does not support that endpoint and to send Jev to POST /v1/systemone. Call it with TypeSafe’s SDK or the /v1/systemone endpoint instead.

The SDK gets a 401 with "invalid JWT header"

Check the base URL. TypeSafe’s SDKs add /v1/systemone themselves, so a base URL ending in /v1 sends the request to /v1/v1/systemone, which isn’t a gateway API path, and the gateway rejects your API key there. Use https://app.barndoor.ai/api/llm-gateway.

I can't add Jev to an existing route

That route holds chat models, and a route can’t mix the two. Create a separate route for Jev.

​Before you start

​Find the name to call

​Call Jev

​Usage and cost

​Troubleshooting

Before you start

Find the name to call

Call Jev

Usage and cost

Troubleshooting