Jev on GitHub
If you're searching GitHub for Jev, there are two honest answers. You won't find the model itself: Jev is TypeSafe AI's closed-weight, hosted System One model — you call it as an API, and there's no source repo, weights, or training code to clone. What you will find is a real and growing shelf of code built around that API: an MCP server, a computer-use safety gate, community SDKs in several languages, benchmark suites, and curated awesome lists. Everything below is community-built and unaffiliated with TypeSafe — here's the map, with the real repos.
Two different things get searched as "jev github": is the Jev model's source or weights on GitHub, and is there any Jev code on GitHub I can actually use. The first is no; the second is a definite yes. This page covers both honestly and links the real repositories — all of them community projects, not TypeSafe's own code.
Is Jev itself on GitHub?
No. Jev is proprietary and hosted — TypeSafe AI serves it behind an API (you send state plus typed questions and get back calibrated, typed answers), and it hasn't published the weights, the training code, the data, or an official source repo. So there is no `typesafe-ai/jev` to clone, no model definition to read, and no checkpoint on a releases page. Any repo with "jev" in its name is a third-party project wrapping or reproducing the API, not the model itself.
That's deliberate, and it's the same reason Jev isn't on Hugging Face: the hard part isn't only the architecture — a non-autoregressive parallel sampler that emits a structured value in one forward pass — it's the RLCD calibration (Reinforcement Learning for Calibrated Decisions) that makes its confidence numbers trustworthy. A repo can copy the request shape and even the architecture, but it can't inherit that calibration.
What is on GitHub — code you can actually use
The ecosystem filled in fast because Jev returns a typed decision in one call, so it drops cleanly into agents, routers and guardrails. These are the repos worth opening first. They're all community projects — not built or endorsed by TypeSafe — so check each one's license yourself:
Give an agent a typed decision tool
# MCP server: exposes a `decide` tool to any MCP client (Claude Code, Claude Desktop, …)
# Runs straight from GitHub over npx — nothing to build:
export JEV_API_KEY=jv_live_...
npx github:codaaiteam/jev-mcpThat registers one decide tool your agent can call whenever it needs a typed, calibrated answer instead of a guess — see the MCP guide on this site for client config. A sister repo, codaaiteam/jev-computer-use, wraps the same idea as a safety gate: before a coding or computer-use agent runs a risky action, a Jev noul call returns a calibrated yes/no on whether to allow it.
Browse the full directory
The community awesome lists are the cleanest way to find SDKs (Go, Ruby, Java, Kotlin, Elixir and more), demos and integrations — none of which are official, so treat each as a separate community project. The benchmark repos let you reproduce the public JevBench task set on your own machine.
- codaaiteam/jev-mcp — Community MCP server — gives any MCP client a typed decide tool over npx. MIT. Unaffiliated with TypeSafe.
- codaaiteam/jev-computer-use — Community gate: a typed, calibrated Jev safety check in front of a coding / computer-use agent's actions. Unaffiliated with TypeSafe.
- hellogumbo/awesome-jev — Community directory of projects built on Jev — SDKs, demos, integrations. CC0. Unaffiliated with TypeSafe.
- cobanov/awesome-jev — Another community-curated, source-backed list of projects built with Jev.
- fstandhartinger/jevbench — Open benchmark for Jev-class typed decision models — run the task set yourself.
Want an open model you can actually clone?
If what you came to GitHub for is weights you can download and run, the Jev model won't give you that — but several open, self-hostable decision models answer the same typed question. They're separate projects, not Jev and not from TypeSafe, and none are RLCD-calibrated, so you validate confidence on your own data. See the open-source and Hugging Face pages for the full map.
Or skip the clone entirely
If you just want to make a decision rather than host one, you don't need any repo. Our hosted endpoint serves the same choice / score / noul API with a jv_live_ key — try it free in the browser playground with no signup, then POST to https://jevtypesafeai.com/api/v1/decide (official upstream https://api.typesafe.ai/v1/systemone). Each decision costs a fraction of a cent and output is free.
FAQ
Is Jev on GitHub?
The Jev model is not — it's TypeSafe AI's closed-weight, hosted System One model, so there's no official source or weights repo to clone. What is on GitHub is the ecosystem around the API: an MCP server, a computer-use gate, community SDKs, awesome lists and benchmark suites. All of those are third-party projects, not TypeSafe's own code.
Where's the Jev source code?
There isn't a public one. TypeSafe keeps Jev's weights, architecture and training code closed, because the hard part is the RLCD calibration that makes its confidence trustworthy — a reimplementation can copy the request shape but not inherit that. You call Jev as a hosted API rather than building it from source.
What Jev repos can I actually use?
Community ones: codaaiteam/jev-mcp (a decide tool for MCP clients, runs over npx), codaaiteam/jev-computer-use (a typed safety gate for agent actions), the awesome-jev directories (hellogumbo, cobanov and others) for SDKs and demos, and fstandhartinger/jevbench to reproduce the benchmark set. None are made by TypeSafe — check each repo's license yourself.
Is there an open-source model I can clone instead?
Not Jev itself, but open Jev-like decision models exist — laya (Apache-2.0), CLM-8B, pplx-decider and others — which you can download and self-host. They accept the same decision shape but aren't RLCD-calibrated, so you validate confidence on your own data. See the is-jev-open-source and jev-huggingface pages.
How do I use Jev without cloning anything?
Run a real decision in the browser playground with no signup, then grab a jv_live_ key and POST to https://jevtypesafeai.com/api/v1/decide (upstream https://api.typesafe.ai/v1/systemone). A decision costs a fraction of a cent and output is free — no repo, no self-hosting.
See also: Jev MCP server · Jev for computer use · Is Jev open source? · Jev on Hugging Face · Awesome Jev · JevBench · Get a key
Nothing to clone — just decide
The Jev model isn't a repo you build; it's a hosted, calibrated API. Run a real typed decision in the browser with no signup, then grab a jv_live_ key. Sub-cent per decision, output free.