OpenJev appeared on GitHub within days of Jev’s launch and answers the question every open-source builder asked immediately: can the System One decision contract be rebuilt without TypeSafe’s private weights? SiliconLabAI’s answer is an elegant approximation — ordinary chat LLMs used as micro-scorers behind Jev’s exact interface: fixed answer spaces, choice/score/yes-no questions, probability distributions and confidence. It is emphatically not real Jev, and says so itself. Here’s how it works, what it’s for, and the caveats that matter.
Short answer
- OpenJev = an open approximation of TypeSafe’s Jev: same decision contract (choice / score / yes-no with probabilities), powered by ordinary LLMs as micro-scorers.
- The trick: a parallel sampler — each option scored independently with a tiny {“p”: 0–1} call, then logit→softmax into a distribution; everything runs via Promise.all.
- Its own honesty: “not a weight-compatible reimplementation” — real Jev is a specialised model at 70–500ms; OpenJev’s example run: 620ms and 9 parallel calls on gpt-4o-mini.
- Status check (21 Sept 2026): ~32 stars, 4 forks, TypeScript, about a day old — and no licence file yet despite the “OpenSource Jev” label.
What OpenJev actually is
The README’s framing is exactly right: instead of one flaky “return a giant JSON blob” call — where models drop keys, invent labels and break JSON — OpenJev decomposes every decision into tiny, constrained micro-calls. For a choice with four options, it fires four parallel prompts of the shape “the correct answer is billing (charges, refunds, invoices)” and asks only for a probability back. Independent probabilities become logits, softmax turns them into a proper distribution, and you get Jev-style output: the pick, per-option probabilities, and a confidence derived from the winner’s margin.
The interface mirrors Jev’s three question types faithfully — choice (argmax plus confidence), score (expected value across your levels, interpolation allowed, so “urgency 2.4 of 3” is a legal answer), and yes/no as a single probability. One POST to its evaluate endpoint carries your state and all questions; everything — options within questions, and the questions themselves — runs in parallel. It’s the same constrain-then-decide philosophy from the Jev architecture guide, rebuilt from parts anyone has.
OpenJev vs real Jev: the honest comparison
| Real Jev | OpenJev | |
|---|---|---|
| Engine | Specialised System One model (RLCD-trained) | Any chat LLM as micro-scorer (gpt-4o-mini, Qwen, local…) |
| Latency | ~70–500ms single pass | ~620ms in its own example (9 parallel calls) |
| Cost shape | $0.042/M input, free output | You pay the scorer model per micro-call — example run: 1,840 input tokens for one decision set |
| Calibration | Trained-for (RLCD) | Improvised via logit/softmax — uncalibrated by construction |
| Weights | Private, hosted only | No weights at all — it’s orchestration code |
Read the cost row twice: on token-billed scorers, OpenJev’s many-tiny-calls design multiplies input tokens — nine calls for one ticket — so it’s not automatically cheap. Where it gets interesting is pointing the base URL at models that cost nothing: a local Qwen behind an OpenAI-compatible endpoint makes the whole contract genuinely free — the full setup logic lives in the run-Jev-locally guide.
🔥 Want this set up without the guesswork? Prototyping decision layers on OpenJev before wiring the real thing 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.
Who should use OpenJev (and the caveats)
Three honest use cases. Learning: it’s the best readable explanation of the System One contract in existence — a few hundred lines of TypeScript instead of a whitepaper. Prototyping: build your schemas and confidence thresholds against OpenJev today, swap the endpoint for real Jev when your waitlist clears — the interface travels. Sovereignty: if hosted-only is a dealbreaker, an approximation you control beats a real thing you can’t run.
Caveats, dated 21 September 2026: the repo is about a day old from an unfamiliar org (SiliconLabAI), and despite the “OpenSource Jev” label there is no licence file yet — legally that’s all-rights-reserved until one lands, so don’t build it into anything commercial before checking. It also takes API keys, so the read-the-code rule applies in full. And its confidence numbers are improvised, not calibrated — treat thresholds tuned on OpenJev as needing re-tuning on real Jev.
The bottom line on OpenJev
OpenJev is the ecosystem working exactly as it should: days after a closed model defined a new contract, an open approximation of that contract exists, honestly labelled and cleverly built. Use it to learn the pattern, prototype your schemas, or go fully local — just carry its own disclaimer with you: this is the shape of Jev, not the speed, the calibration, or (until a licence file appears) even formally the openness. The real thing’s routes are in the API guide.
FAQ: openjev
What is OpenJev?
An open-source-styled approximation of TypeSafe’s Jev: the same decision contract (choice/score/yes-no with probabilities and confidence) rebuilt using ordinary chat LLMs as parallel micro-scorers.
Is OpenJev the same as Jev?
No — its own README says it: real Jev is a specialised RLCD-trained model at 70–500ms; OpenJev is an architectural approximation, not weight-compatible.
How does OpenJev work?
Each option gets a tiny independent “probability this is correct” call; results go logit→softmax into a distribution; questions and options all run in parallel.
Is OpenJev actually open source?
The code is public, but at the time of writing there’s no licence file — formally all-rights-reserved until one is added. Check before commercial use.
Is OpenJev cheaper than Jev?
Not necessarily — you pay your scorer model per micro-call (nine calls for one example decision set). It’s genuinely free only when pointed at local models.
What’s OpenJev best for?
Learning the System One pattern, prototyping schemas you’ll later run on real Jev, and fully-local decision engines where hosted-only is a dealbreaker.
Next step: if you want the decision contract running on infrastructure you control 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 GitHub: Inside SiliconLabAI’s Day-Old Repo
- Is Jev Open Source? No — But The Ecosystem Is
- Jev Architecture: How System One Actually Works
Last updated September 2026. This is the living guide to openjev — it gets updated as the tools change.