Blog / Questions / FIG. 60
Is Jev Open Source? No: Here's the Actual Landscape
Jev's weights are not open source. What is open around it (SDKs, tooling), the open-source alternative TypeLLM, and when self-hosting actually pencils out.
No: Jev is a hosted, proprietary model from TypeSafe AI, accessed through their platform. The weights aren't published, there's nothing to download, and "can I run it myself" has a one-word answer.
What is open is most of the scaffolding around it. The community client libraries (jev-harness in TypeScript, jevclient in Python, req_llm's Elixir provider, swift-typesafe) are open repositories, as is tooling like jev-tree and the awesome-jev ecosystem list. So the integration layer is inspectable even though the model isn't, which matters for the security review your data-handling questions will raise.
For the thing people usually mean by this question ("I want the capability without the vendor"), the ecosystem has a named answer: TypeLLM, an open-source project bringing the same typed-decision interface (closed questions, structured answers) to models you host yourself. The trade is the standard one: per-verdict pricing and zero infrastructure versus tuning work, serving costs, and control. At reported Jev prices, self-hosting pencils out mainly at extreme sustained volume, for data-residency requirements the hosted path can't meet, or on principle, which is also a valid budget line.
The middle path most teams actually land on: hosted Jev for speed-to-working, with the open typed-decision pattern as insurance that your question sets, the real asset, port anywhere. Questions are yours; weights were never going to be.
Frequently asked questions
Will TypeSafe open-source Jev later?
No public commitment exists either way at the time of writing; plan on hosted-only and treat anything else as upside.
Is there an open-source model that does what Jev does?
TypeLLM implements the typed-decision interface over self-hosted open models; performance depends on the model you put behind it. The alternatives page frames the decision.
Does closed-source matter for my use case?
It matters for data residency, auditability requirements, and vendor risk; it doesn't change the architecture, since cascades and question design are model-portable by construction.
Numbers throughout are as reported by the build authors, not verified by shipwithjev. Code-shaped examples are pseudocode; the official docs live at docs.typesafe.ai.