Running Jev locally

If you are searching for Jev local, the honest answer is that you cannot run Jev locally: it is a hosted API with no public weights, and there is no supported way to self-host the model.

There is no Jev local build. TypeSafe AI ships Jev as a hosted API, and the model runs only on their servers. That means there are no open or downloadable weights, no offline binary, and no official way to run Jev locally or self-host it. If you have seen the phrase "run Jev locally" in a tutorial, it is a misunderstanding — a Jev local deployment simply is not something TypeSafe AI offers today.

Why there is no Jev local option

Jev is a System One model that returns typed, calibrated decisions, and it stays on TypeSafe AI's infrastructure. What is open is the tooling around it — projects like awesome-jev, jev-mcp, and the LangChain integration — but that tooling still calls the hosted API. So even with the open ecosystem, running Jev locally means the model itself is never on your machine; you are only wiring up clients to a remote endpoint.

Hosted ways to access Jev

Instead of a Jev local install, use one of the hosted paths. Each keeps the model remote but gives you the same typed decisions in a single round trip.

• Official API: POST https://api.typesafe.ai/v1/systemone, direct from TypeSafe AI
• Our hosted resale: POST https://jevtypesafeai.com/api/v1/decide with a jv_live_ key (get one at /pricing, no waitlist)
• Gateways and clients: the open tooling around Jev routes your calls to the hosted model

Keeping latency low without Jev local

People want to run Jev locally mainly for speed, but the hosted API is already fast — roughly 70–500ms per call, about 40–200x faster than an LLM, at around $0.0004 per decision. To keep latency low without a Jev local setup, batch related questions into one call so they answer in parallel, deploy your service near the endpoint, reuse connections, and pin a model version so behavior stays predictable. In short, there is no Jev local mode, but the hosted routes give you typed, calibrated decisions fast enough that a Jev local copy is not needed.

See also: Get Jev · How to use the API · Pricing

Related: Jev API key · Jev architecture · Jev vs an LLM