Jev API access: five ways to call it

Five routes are documented today: TypeSafe's own API, OpenRouter, Vercel AI Gateway, Netlify AI Gateway and OpenCode Zen. All five send the same kind of request and return the same kind of answer. What changes from route to route is the model name, the web address you call, the key you use and the company that bills you.

Jev itself is TypeSafe's decision model: it reads some text and returns a typed answer with a probability, not a piece of writing. If you are new to it, start with how Jev works, then come back here to pick a route.

• How to use Jev: your first API call

Every request has a model name, a state (the text or facts Jev should look at) and a questions object. Each question has a type, which is choice (pick one option), score (rate on a scale you define) or noul (a yes or no probability from 0 to 1), plus the instructions for that question. The answer comes back with a probability for each option and a confidence value.

This example calls TypeSafe's own endpoint with one yes or no question. It is assembled from TypeSafe's quick start and Vercel's published example. Put your key in TYPESAFE_API_KEY first.

curl https://api.typesafe.ai/v1/systemone -H "Authorization: Bearer $TYPESAFE_API_KEY" -H "Content-Type: application/json" -d '{"model":"jev-latest","state":"The agent fixed the checkout bug and all tests pass.","questions":{"continueWorking":{"type":"noul","instructions":"Should the agent take another step?"}}}'

TypeSafe's quick start tells you to create a key in its console at console.typesafe.ai/keys and keep it in an environment variable named TYPESAFE_API_KEY. You then call POST https://api.typesafe.ai/v1/systemone with the key as a Bearer token. Two official SDKs wrap the call. Both install with one command:

npm install @typesafe-ai/sdk

python3 -m venv .venv

.venv/bin/pip install typesafe-sdk

On this route the name jev-latest points to jev-1.14.0 and is the default in the SDKs, according to TypeSafe's models page. Aliases can move when a new version ships, so pin a version such as jev-1.14.0 if you want results that never change under you. A request can carry 64,000 tokens in total and 32,000 per question, and an opt-in header raises both to 262,000. Rate limits are set per organisation and answered with a 429 error when you exceed them.

A TypeSafe key needs credit before it can be created, which is the hurdle most people hit first. The access guide has the details and the free ways to try Jev by hand.

• Can you use Jev for free?

OpenRouter lists three Jev model IDs: typesafe/jev-1.14, typesafe/jev-1.13 and the alias ~typesafe/jev-latest. You need an OpenRouter key and nothing else: its page says there is no separate TypeSafe account or waitlist, and usage is billed to your OpenRouter account. Responses include a usage.cost field with the cost in US dollars.

There are two ways in. The Decisions API is POST https://openrouter.ai/api/alpha/decisions. If you already use TypeSafe's SDK, you can point it at https://openrouter.ai/api/v1/systemone instead and change nothing else. OpenRouter gives a context length of 262,144 tokens, but warns that requests with more than about 32,000 bytes of text, and every request with an image, run in a slower, less accurate long-context mode. Box questions need typesafe/jev-1.14 or the latest alias, because Jev 1.13 answers them with a 400 error.

The page does not give a dollar price. It says you pay per input token at the price on the Jev model page, with output free.

• Jev on OpenRouter: model IDs and the Decisions API

On Vercel the model ID is typesafe-ai/jev. Vercel's model page lists $0.042 per million input tokens, a context window of 32,000 tokens and two providers, typesafe-ai and DigitalOcean, with no free credit. You use an AI Gateway key, and every route below is billed through Vercel.

Vercel's changelog of 21 September 2026 describes three ways in. With TypeSafe's own SDK you change the base URL to https://ai-gateway.vercel.sh/typesafe and the key to your gateway key. With plain HTTP you send a Bearer-token request to https://ai-gateway.vercel.sh/v1/evaluate. Or you use the experimental_decide function in Vercel's AI SDK. Vercel's HTTP example names the yes or no type boolean, where TypeSafe's own API calls it noul, so use the names from the page you are following.

Netlify is the simplest route if you already deploy there. Its changelog entry for Jev 1.14.0, last updated on 24 October 2026, says Jev 1.14.0 works through Netlify's AI Gateway with zero configuration, in Netlify Functions and in its scheduled, background and edge functions. You install @typesafe-ai/sdk, create new TypeSafeClient() and call systemOne with the model jev-1.14.0. No key, provider setting or base URL is needed because the gateway supplies the credentials, and usage is billed to your Netlify credits.

Netlify's first entry, from September 2026, requires Node.js 20 or newer. Neither page states a price or a usage limit, so check your Netlify credit use after your first few runs.

Yes, on Jev 1.13 only. OpenCode's Zen page lists jev-1.13, at $0.042 per million input tokens with free output, and jev-1.13-free, free for a limited time with no end date given. The page documents a direct call to https://opencode.ai/inference/systemone/v1/systemone with a Zen key, in the same request format as TypeSafe's API. It does not say how to pick Jev inside the OpenCode terminal app.

• Jev in OpenCode: skill, approvals and Zen

Our read, from the pages above:

• Already deploying on Netlify or Vercel: use that gateway. Netlify needs no key at all, and both bill Jev beside the rest of your usage.
• Already paying OpenRouter: use it. One key and one bill, and it is the only gateway here that lists both Jev 1.14 and 1.13 under named model IDs.
• Want the newest version, the SDKs and TypeSafe's own limits: call TypeSafe directly. This is the only route whose behaviour TypeSafe's own pages describe in full, and the only one where a key needs credit first.
• Just want to try a call from your own code: OpenCode Zen's free Jev 1.13 model, while it lasts and with data you would not mind sharing.
• Planning a large volume: compare the prices on each page and run your numbers in the cost calculator before you commit.

• Jev cost calculator

The same request needs small changes on each route, and these are the usual snags:

• The model name is different everywhere. jev-1.14.0, typesafe/jev-1.14, typesafe-ai/jev and jev-1.13 all mean Jev, on different services.
• The version may not be the one you expect. OpenCode Zen runs 1.13 only, Netlify names 1.14.0, and Vercel's typesafe-ai/jev does not state a version on its page.
• Image and box questions are early features. Box needs Jev 1.14, and OpenRouter warns that images use the slower long-context mode. TypeSafe labels both as not ready for production.
• Limits differ. TypeSafe allows 64,000 tokens per request, Vercel's page says 32,000, and OpenRouter says 262,144. Check the limit on the route you use.
• A gateway's page can lag behind TypeSafe. Netlify's first entry still described jev-latest as 1.13.0 when we read it on 25 October, after TypeSafe's models page had moved it to 1.14.0, so pin the version if you need to be sure.