User repository on GitHub
- Forks
- 0
- Last pushed
- Updated Sep 2026
Repo stats from the GitHub API, cached Oct 2026.
CommunityJev Tools & CLIs
JevSysUno
Reference TypeScript integration for Jev that keeps thresholds in your code and confidence bands in the type: a decision builder, runtime response validation, a rule fallback, and an eval CLI that refuses thresholds from thin data.
- Stars
- 0
- Language
- TypeScript
- Published by
- Community
- Author
- Dujaydis
- Last verified
- Oct 2026
About this project
A reference integration for TypeSafe AI's Jev, the first "System One" model: typed decisions, thresholds that live in your code, and a calibration harness that tells you whether to trust any of it.
Quoted from the project's README. Read the full README.
Works forJev for Classification
Used inJev for Software & DevTools
- Languages
- TypeScript
Highlights
- The result union has no `outcome` at its root, so reading one requires narrowing on `kind` first.
- `npm test` runs 65 tests with no API key, and `npm run eval` produces a simulated report without one either.
- The eval report breaks accuracy out by difficulty tag and computes ECE, Brier, AUROC and reliability bins against your baseline.
- Errors split into degrade and fail: transient problems fall back to rules, but a 401 or 422 rethrows instead of hiding a config bug.
Watch out
Nothing in this repository has been run against the real model. The status line states no live API access was available, that the wire contract was taken from TypeSafe's official SDK source, and that the benchmark figures come from independent third-party evaluations. Those benchmarks disagree with each other and are reported as found: on a phishing set Jev scored 62.6% against Claude Haiku 4.5 at 81.3% while a plain regex scored 91.8%, and 2.2% of identical repeated inputs returned a different label. The README's conclusion is that Jev's edge is latency, cost and an interface that cannot break, not accuracy. Cost estimation returns `undefined` by default because the widely-quoted per-million-token input figure is not confirmable from a first-party source, and production use requires pinning a concrete model name since `jev-latest` moves. Copy reviewed against the repository README on 2026-10-04, not from running it.
Questions about this entry
Short answers drawn from the linked project or page.
Not first-party evidence, and the README is upfront that the numbers come from independent third-party evaluations while the wire contract comes from TypeSafe's SDK source. What makes it worth reading is that the results are unflattering and left in place: on a phishing set Jev scored 62.6% against Claude Haiku 4.5 at 81.3%, a plain regex beat Jev's best question at 91.8%, and 2.2% of identical repeated inputs returned a different label. The stated conclusion is that Jev's edge is latency, cost and an interface that cannot break.
Not through the types, which is the point of the design. The result union deliberately has no `outcome` at its root, so you cannot read one without narrowing on `kind` first, and the kinds are acted, review, escalated, fallback and failed. Low confidence is also never treated as a fallback reason: the README is explicit that it is a successful answer belonging in the escalate band, so a rules fallback cannot quietly absorb a model that merely is not sure.
From data, not from the placeholders. The shipped example uses act at 0.9 and review at 0.6, and the README says to replace them with values from an eval report on your own labeled data while keeping the provenance comment. That report breaks accuracy out by difficulty tag and computes ECE, Brier, AUROC and reliability bins, and it refuses to suggest thresholds from fewer than 50 labeled examples. `npm run eval -- decisions/ticket-triage --simulate=overconfident` runs without an API key.
It throws rather than degrading quietly. The README notes the JavaScript SDK's `parseBody` casts instead of validating, since the Python one uses pydantic, so without the project's `validate.ts` a vendor change flows straight into your decision code typed as something it isn't. A contract break raises instead of falling through to the rules fallback. Errors are also split: transient failures degrade to the fallback, while 401 and 422 rethrow, because a config bug that looks like a working degraded service surfaces weeks later as a quality regression.
No. npm test runs 65 tests without one, and npm run eval -- decisions/ticket-triage produces a simulated report without one too; only the real API path needs a key.
It breaks accuracy out by difficulty tag, computes ECE, Brier, AUROC, and reliability bins, compares Jev against your deterministic baseline, and refuses to suggest thresholds from fewer than 50 labeled examples.
defineDecision<State>()({...}) is curried so the state type and the inferred question literal both survive; it uses act and review thresholds you set, and its decide function turns typed answers into your outcome while a fallback handles Jev being down.
More Jev resources like this
38GitHub stars
Research dashboard for Polymarket BTC five-minute Up/Down rounds that feeds market state to Jev class scores and paper-trades the edge.
Jev Tools & CLIs#community#jevbest#catalog-import
38GitHub stars
Python memory filter that judges every stored memory against the current request with one calibrated yes/no per memory in a single Jev request, keeping everything above a threshold.
Jev Tools & CLIs#community#jevbest#catalog-import
Community
samdotmak/jev-recallTypeScript
samdotmak/jev-recallTypeScript30GitHub stars
Browser extension that filters your X timeline with a model you choose and pay for, using plain-language rules plus per-author and blocked-word controls.
Jev Tools & CLIs#community#jevbest#catalog-import
Community
tshmieldev/sharpTypeScript
tshmieldev/sharpTypeScript27GitHub stars
Pi extension that replaces many small tool calls with one typed TypeScript execute_tools call, type-checked before running in a sandbox.
Jev Tools & CLIs#community#jevbest#catalog-import
Community
boozedog/pi-codemodeTypeScript
boozedog/pi-codemodeTypeScript17GitHub stars
Pi extension that trims wasted context before each model call by deduplicating repeated tool results, stripping failed payload args, and letting the model summarize closed spans.
Jev Tools & CLIs#community#jevbest#catalog-import
Community
Davidcreador/pi-dcpTypeScript
Davidcreador/pi-dcpTypeScript18GitHub stars
Turkish Next.js app that classifies transcript passages with Jev instead of rewriting them, finding chapters, heatmap moments, claim types and topic matches while showing original wording.
Jev Tools & CLIs#community#jevbest#catalog-import
Community
sensahin/transcript-lensTypeScript
sensahin/transcript-lensTypeScriptFrom the Jev guides
Original write-ups on what this entry is for.
The request shape barely changes across surfaces; the surrounding contract does. When to call the API directly and when a gateway, a Worker binding, or a router entry earns its extra hop.
Jev Tools & CLIs#gateway#vercel#cloudflare
Read article
Taking a decision checkpoint out of Python: what the GGUF and browser WebGPU exports preserve, what they change, and the parity evidence the model cards publish.
Jev Tools & CLIs#open-models#local#onnx
Read article
Jev labels a record that already exists. Code moves, flags, or holds it. How the document, mail, and transfer tools keep the action out of the model.
Jev Tools & CLIs#classification#email#security
Read article
From the Jev community
Posts from builders shipping with Jev right now.