Is Jev open source? What is open and what is not

No, not the model. Jev is closed: TypeSafe, the company that makes it, has not published its weights (the learned numbers that make up the model), its size or its architecture. You cannot download Jev, run it on your own computer, look inside it or train it further. You use it through TypeSafe's API, and a few gateways resell that same access.

What is open is the tooling around it. TypeSafe publishes its Python and JavaScript SDKs, its agent skills, an n8n node and some examples as open source, under the MIT licence. They help you call Jev; none of them contains Jev.

• What is Jev?
• Jev API access: five ways to call it

TypeSafe's own pages say nothing about weights or a licence. Its launch post describes "a new model architecture" and a training method it calls RLCD without detail, and its documentation says only that Jev is not trained to generate text. We then checked where weights would be published, on 28 October 2026:

• GitHub. TypeSafe's organisation has repositories for SDKs, skills, an adapter, an n8n node, evaluation code and examples, but no repository for the model.
• Hugging Face. Two accounts are named TypeSafe AI (typesafeAI and typesafe). Neither is marked verified, so we cannot confirm they are official, and both list no models. A search for "jev" returns models made by other people in Jev's style.
• What this does not prove. It is a check for something missing, not a statement from TypeSafe. The company has not said whether it will ever open the model; its launch post says only "we have a lot more in the pipeline".

Several things TypeSafe publishes are open source, each under its own licence in its own repository:

• SDKs (MIT). The Python and JavaScript libraries that send requests to Jev. You can read and change them, but they still need a TypeSafe or gateway key to get an answer.
• Agent skills (MIT). The files that teach coding agents such as Claude Code and Codex to write programs that call Jev.
• An n8n node, an adapter and examples (MIT). Ready-made pieces for automations and small projects.
• Workflow evaluation code (Apache 2.0). A repository named WorkflowEvals.

• Jev in Claude Code, Codex, OpenCode and Cursor

Because Jev is closed, four things follow:

• Your text goes to someone else's servers. Every request is sent to TypeSafe's API or to the gateway you use. You cannot keep the data inside your own network.
• You cannot inspect or change the model. If Jev gets something wrong, you can change your questions and thresholds, but not the model.
• You depend on one company's prices and rules. TypeSafe's models page lists $0.042 per million input tokens with free output today, and prices can change.
• You can pin a version. Sending a version ID such as jev-1.14.0 instead of the alias jev-latest keeps answers from changing when a new release ships.

• Jev pricing
• Can you use Jev for free?

If you need a model you can run yourself, several open decision models do a similar job. The three below are compared with Jev in full on this site. They differ in size, speed and accuracy, and the numbers in those comparisons are each company's own unless an article says otherwise.

• Jev vs open models: can you skip Jev?
• Jev vs Laya: pay per call or run it yourself?

Based on what is published today:

• Use Jev if you want a hosted service with public pricing, no hardware to run and the tooling above.
• Use an open model if the data must stay on your own machines, you need to inspect or fine-tune the model, or you want a fixed cost on hardware you own.
• Test both on your own data before you decide: run the same labelled sample through each and compare accuracy, speed and total cost.