Can you run Jev locally? The straight answer: no — not the real one. TypeSafe’s model is hosted-API only, no weights have been released, and nothing about that has changed as of 21 September 2026. But the question deserves the fuller answer, because a genuine Plan B exists: OpenJev, the community’s architectural approximation, runs entirely on your machine and can point at local models — giving you the System One decision contract, if not the model, with zero cloud involved. Here’s the honest map.
Short answer
- Real Jev locally: no — hosted API only (waitlist or Vercel Gateway); weights private; even OpenJev calls itself “not weight-compatible”.
- Plan B: OpenJev runs locally (npm install, npm run dev, localhost:3001) and its base-URL field accepts any OpenAI-compatible endpoint — including local models.
- Fully-local recipe: OpenJev + a local model (its own examples list Qwen-class models) = the decision contract at genuinely $0, offline-capable.
- Trade-offs: slower than real Jev (~620ms example even on hosted gpt-4o-mini; local adds more), improvised rather than trained calibration.
Can you run Jev locally? What’s actually possible
Separate the two things people mean. The model: TypeSafe serves Jev exclusively through its API — there’s no download, no parameter disclosure, no self-host option, and the 70–500ms magic includes their custom parallel sampler on their hardware. The contract: schema in, typed decisions with probabilities out — and THAT is reproducible at home, because OpenJev rebuilds it in TypeScript using any chat model as a micro-scorer. Three commands (npm install, npm run dev, open localhost:3001) and the decision engine is running on your Mac.
The local unlock is OpenJev’s base-URL field: it defaults to OpenAI, but accepts any compatible endpoint — which is exactly the shape local model servers expose. Point it at a local Qwen-class model (its own model examples include one) and every micro-scoring call stays on your machine: no cloud, no per-token bill, no data leaving the room.
The fully-local recipe and its honest trade-offs
| Layer | Hosted real Jev | Fully-local Plan B |
|---|---|---|
| Decision engine | TypeSafe’s API | OpenJev on localhost:3001 |
| The scorer | Jev itself (RLCD-trained) | Your local model via base URL |
| Latency | ~70–500ms | 620ms+ hosted example; local models add more |
| Cost | $0.042/M input (free on Vercel to 25 Sept) | Genuinely $0 after setup |
| Privacy | State goes to the API | Nothing leaves your machine |
| Calibration | Trained-for | Logit/softmax improvisation — re-tune thresholds |
The privacy row is Plan B’s real argument: decision workloads are often exactly the data you’d hesitate to ship out — support tickets, leads, internal documents. A local decision layer judging local data with a local model is the same sovereignty logic as the local Hermes stack, applied to the System One layer. The speed row is the price: reflex-fast becomes merely fast, and ten-a-second use cases (the Doom demo class) stop being realistic.
🔥 Want this set up without the guesswork? A local-first decision layer for data that can’t leave the building is exactly the kind of thing we set up together inside the AI Profit Boardroom — 3,700+ members, four live calls a week, daily tutorials, done-for-you templates and a 30-day roadmap. Prefer 1-on-1 help? Book a free SEO strategy session and we’ll map it out for your business.
Choosing your lane
My honest routing: if you’re piloting capabilities this week, use the real thing — it’s free on Vercel until the 25th and you’ll learn what good looks like. If you’re prototyping schemas for later, either lane works — the contract is identical by design. Go fully local when privacy or offline operation is the requirement rather than a preference — and accept that you’re running an approximation whose confidence numbers deserve scepticism until you’ve tested them on labelled cases.
Honesty box: OpenJev is about a day old from an unfamiliar org, currently ships without a licence file, and takes API keys — read the code first, and don’t commercialise on it until licensing lands. And if TypeSafe ever releases weights or an on-prem option, this page changes completely; as of 21 September 2026 there’s no sign of either.
The bottom line on can you run Jev locally
Can you run Jev locally? The model, no — and probably not soon. The contract, yes — OpenJev plus a local model puts schema-locked decisions with confidence scores entirely on your own hardware, at zero cost, with real trade-offs in speed and calibration. Hosted for capability, local for sovereignty — and the schemas you write travel between both.
FAQ: can you run jev locally
Can you run Jev locally?
Not the real model — it’s hosted-API only with private weights. OpenJev, the open approximation, runs locally and can use local models as scorers.
How do I run the local Plan B?
npm install and npm run dev on OpenJev (localhost:3001), then point its base-URL field at a local OpenAI-compatible model endpoint.
Is the local route really free?
After setup, yes — local scorer models have no per-token cost. The trade is speed and improvised calibration.
Is local Jev as fast as real Jev?
No — real Jev cites 70–500ms; OpenJev’s own hosted example runs ~620ms, and local models add more.
Why go local at all?
Privacy and sovereignty: tickets, leads and internal documents get judged without leaving your machine.
Will TypeSafe release weights?
Nothing announced — as of 21 September 2026 there’s no download or on-prem option, and no stated plans for one.
Next step: if you want the right lane — hosted or local — for your decisions working for you this week, join the AI Profit Boardroom for the full walkthroughs and live help — or book a free SEO strategy session and I’ll point you at the fastest path for your situation.
About Julian Goldie: SEO agency owner with 10+ years in SEO, 394K+ subscribers on YouTube, a 100% job-success score on Upwork, 75K+ members across his communities, and author of a best-selling SEO book. He runs the AI Profit Boardroom community and offers a free SEO strategy session.
Related reading
- OpenJev: The Open-Source Jev Approximation
- Is Jev Open Source? No — But The Ecosystem Is
- Hermes Desktop Local: Run It All On Your Machine
Last updated September 2026. This is the living guide to can you run jev locally — it gets updated as the tools change.