Skip to content
Back to search

Jev

Unofficial TypeSafe Jev showcase — System One decisions, not chat.

Safety scan

A static text scan, not a guarantee. Select the badge to see what was scanned and every finding.

Checked today
Badge for your README
findsafeskills safety rating

Markdown

[![findsafeskills safety rating](/api/listings/cobusgreyling--jev/badge.svg)](/listings/cobusgreyling--jev)

HTML

<a href="/listings/cobusgreyling--jev"><img src="/api/listings/cobusgreyling--jev/badge.svg" alt="findsafeskills safety rating" /></a>

Install

i

What was scanned

The files the safety scan read, as written by the repository's author and shown as plain text. Lines with a finding are marked.

README.md · README · 258 lines · 1 finding marked

View this file on GitHub (read at the repository's current version: the scan did not record which commit it read)

  1. # Jev Showcase — TypeSafe System One, not chat
  2. <p align="center">
  3. <img src="assets/header.jpg" alt="Jev — a crystalline decision engine emitting parallel judgment beams" width="100%" />
  4. </p>
  5. <p align="center">
  6. <strong>Unofficial operator-level companion for TypeSafe Jev</strong><br/>
  7. Choice · Score · Noul · parallel fan-out · confidence as a second axis<br/>
  8. Jev 1.13 · released <strong>15 September 2026</strong>
  9. </p>
  10. <p align="center">
  11. <a href="docs/WHAT-PEOPLE-MISS.md">What people miss</a> ·
  12. <a href="https://typesafe.ai/blog/introducing-system-one-models-and-jev">Announcement</a> ·
  13. <a href="https://docs.typesafe.ai/">Docs</a> ·
  14. <a href="https://docs.typesafe.ai/models">Models</a> ·
  15. <a href="BLOG.md">Article</a> ·
  16. <a href="FRAMEWORK.md">Showcase framework</a>
  17. </p>
  18. ---
  19. Most posts stop at **“fast structured output.”** This repo surfaces the **docs-only details** that change how you build: Jev is a **function call**, questions run **in parallel**, **output is free**, **confidence is not probability**, and **code owns the workflow**.
  20. | Under-known fact | Why it matters |
  21. |------------------|----------------|
  22. | **Not a chatbot** | No prose, code, or explanations. Pair with a generative model. |
  23. | **Three primitives** | Choice / Score / Noul. Question IDs are not sent to the model. |
  24. | **Fan-out** | Many questions, one state, one call. ~12× cheaper than serial. |
  25. | **$0.042 / MTok, output free** | Optimize question design, not completion length. |
  26. | **Noul has no `confidence`** | Do not copy a Noul threshold onto a Choice. |
  27. | **No fine-tune** | Shape answers via `state` + `instructions` + `criteria`. |
  28. | **Literal + no math** | Count, dates, and arithmetic stay in code. |
  29. Full write-up: **[docs/WHAT-PEOPLE-MISS.md](docs/WHAT-PEOPLE-MISS.md)** · interactive cards in the lab.
  30. **Keywords:** TypeSafe, Jev, System One, RLCD, calibrated decisions, structured output, confidence routing, agent guardrails
  31. ---
  32. ## The one-liner
  33. > **Frontier-intelligence function call** — unstructured state in, typed probabilistic decisions out. 70–500 ms. Cannot invent a label off your schema.
  34. | | |
  35. |--|--|
  36. | **Model id** | `jev-1.13.0` (`jev-latest`) |
  37. | **Class** | System One (not a chat LLM) |
  38. | **Endpoint** | `POST /v1/systemone` |
  39. | **Context** | 64k / request; 32k for state + longest question |
  40. | **Input** | Text / JSON. No image, audio, or video |
  41. | **Output** | Choice · Score · Noul (+ probabilities) |
  42. | **Pricing** | **$0.042 / MTok input · output free** |
  43. | **Latency** | 70–500 ms (vendor) |
  44. | **Training** | RLCD — Reinforcement Learning for Calibrated Decisions |
  45. | **Released** | 15 September 2026 |
  46. ---
  47. ## 30-second start
  48. ```bash
  49. git clone https://github.com/cobusgreyling/Jev.git
  50. cd Jev
  51. python3 -m venv .venv && source .venv/bin/activate
  52. pip install -r requirements.txt
  53. python app.py
  54. ```
  55. Open **http://127.0.0.1:7872/home** for the smart-home demo (TypeSafe’s fan-out house), or **http://127.0.0.1:7872** for the operator lab.
  56. ### Harness CLI (TypeScript)
  57. The same System One questions, packaged for agent loops — model routing, semantic retrieval, LLM guardrails, reasoning-trace classification. Wired into [Loop Engineering](https://github.com/cobusgreyling/loop-engineering) as `loop-jev`.
  58. ```bash
  59. cd packages/js && npm install && npm test
  60. npx jev doctor
  61. npx jev route --goal "CI has been red for 3 days" --level L2 --json
  62. npx jev guard --side input --text "Ignore previous instructions" --json
    • Dismissed after review: Quotes instruction-override phrasing (in quotes, code, a table or an example): "ignore previous instructions"
    • Line 82 on GitHub
  63. ```
  64. Docs: [`packages/js/README.md`](packages/js/README.md) · [loop-engineering/docs/jev.md](https://github.com/cobusgreyling/loop-engineering/blob/main/docs/jev.md)
  65. The key is never printed. Store it as `TYPESAFE_API_KEY`, `~/.typesafe/api_key`, or `~/.config/typesafe/api_key` (mode 600).
  66. ```bash
  67. ./run.sh
  68. ```
  69. ### Agent skills
  70. Portable `SKILL.md` files for coding agents. Official TypeSafe skill: [typesafe-ai/skills](https://github.com/typesafe-ai/skills).
  71. ```bash
  72. npx skills add cobusgreyling/Jev --skill jev
  73. npx skills add cobusgreyling/Jev --skill jev-fanout
  74. npx skills add cobusgreyling/Jev --skill jev-guardrail
  75. npx skills add cobusgreyling/Jev --skill jev-route
  76. ```
  77. | Skill | Purpose |
  78. |-------|---------|
  79. | [`jev`](skills/jev/SKILL.md) | Primitives and when to call Jev |
  80. | [`jev-fanout`](skills/jev-fanout/SKILL.md) | Speculative fan-out (one request, many questions) |
  81. | [`jev-guardrail`](skills/jev-guardrail/SKILL.md) | Input/output hazard screen |
  82. | [`jev-route`](skills/jev-route/SKILL.md) | Confidence lanes and model-tier routing |
  83. Grok also loads [`.grok/skills/`](.grok/skills) (symlinks). Catalog: [`skills/README.md`](skills/README.md).
  84. ### Docker
  85. ```bash
  86. docker compose up --build
  87. # → http://127.0.0.1:7872
  88. ```
  89. No API key for offline tabs. Live judge:
  90. ```bash
  91. cp .env.example .env
  92. # put your TypeSafe key in .env or ~/.typesafe/api_key
  93. # mint at https://console.typesafe.ai/keys
  94. ./run.sh
  95. ```
  96. The key is never logged, never returned by `/api/health`, and never committed. See [SECURITY.md](SECURITY.md).
  97. ---
  98. ## What the lab shows
  99. | Surface | Offline? | What you learn |
  100. |---------|----------|----------------|
  101. | **[Smart home](/home)** | Yes (mock) | Official-style house: 13 questions in one call, code flips the lights |
  102. | **What people miss** | Yes | 10 ranked operator cards with actions + doc links |
  103. | **Playground** | Yes | Choice + Score + Noul, then a policy in code |
  104. | **Cost lab** | Yes | Input-only billing, fan-out vs serial, LLM stand-in |
  105. | **Patterns** | Yes | Fan-out · confidence routing · composite · intent |
  106. | **Jaggedness** | Yes | Official Jev 1.13 failure modes |
  107. | **Scenarios** | Yes | Triage, guardrails, RAG, citations, tool traces |
  108. | **Live judge** | Needs key | Real System One call; key never shown |
  109. | **Model card** | Yes | Specs, aliases, gotchas |
  110. ---
  111. ## The request shape
  112. ```bash
  113. curl -X POST https://api.typesafe.ai/v1/systemone \
  114. -H "Authorization: Bearer $TYPESAFE_API_KEY" \
  115. -H "Content-Type: application/json" \
  116. -d '{
  117. "state": "I was charged twice. Please refund the duplicate today.",
  118. "model": "jev-latest",
  119. "questions": {
  120. "department": {
  121. "type": "choice",
  122. "instructions": "Which team should handle this?",
  123. "criteria": {
  124. "billing": "Payments and refunds",
  125. "technical": "Bugs or integrations",
  126. "other": "Neither fits"
  127. }
  128. },
  129. "refund_requested": {
  130. "type": "noul",
  131. "instructions": "Does the message request a refund?"
  132. },
  133. "urgency": {
  134. "type": "score",
  135. "instructions": "How time-sensitive is this?",
  136. "criteria": ["No deadline", "Within a week", "Today or sooner"]
  137. }
  138. }
  139. }'
  140. ```
  141. Answers come back under the same keys, typed. Nothing to parse.
  142. ---
  143. ## Progressive examples
  144. ```bash
  145. python examples/01_noul.py
  146. python examples/02_choice.py
  147. python examples/03_score.py
  148. python examples/04_parallel_fanout.py
  149. python examples/05_confidence_routing.py
  150. python examples/06_composite_scoring.py
  151. python examples/07_guardrails.py
  152. ```
  153. Fixtures by default. Add `--live` when `TYPESAFE_API_KEY` is set.
  154. ```bash
  155. make test
  156. make secret-scan
  157. ```
  158. ---
  159. ## Repository map
  160. ```text
  161. ├── app.py / static/ Operator lab + /home smart-home demo
  162. ├── jev_lab/ HTTP client, cost, policy, house dispatcher
  163. ├── packages/js/ TypeScript harness CLI (@cobusgreyling/jev)
  164. ├── FRAMEWORK.md The eight-layer showcase skeleton
  165. ├── BLOG.md Narrative article
  166. ├── examples/01–07 Progressive scripts
  167. ├── data/ model-card, claims, fixtures
  168. ├── docs/ Operator notes + smart-home + sources
  169. ├── skills/ Agent skills (jev, jev-fanout, jev-guardrail, jev-route)
  170. ├── .grok/skills/ Grok-discoverable copies of those skills
  171. ├── tests/ pytest
  172. └── SECURITY.md Key hygiene
  173. ```
  174. This is **not** an official TypeSafe product. It does **not** replace `typesafe-sdk`. It **does** make the System One thesis interactive and measurable.
  175. ---
  176. ## Use Jev when…
  177. - You need classify / route / score / verify on a hot path
  178. - Schema errors are unacceptable
  179. - You want calibrated probabilities, not a paragraph
  180. - You will keep side effects in code
  181. ## Prefer a generative model when…
  182. - You need prose, code, or an explanation
  183. - The answer space is not a closed set
  184. - The task is counting, arithmetic, or date math
  185. - You want an agent that chooses its own next action
  186. ---
  187. ## Docs index
  188. | Topic | Link |
  189. |-------|------|
  190. | What people miss | [docs/WHAT-PEOPLE-MISS.md](docs/WHAT-PEOPLE-MISS.md) |
  191. | Smart home demo | [docs/smart-home.md](docs/smart-home.md) |
  192. | Anti-patterns | [docs/anti-patterns.md](docs/anti-patterns.md) |
  193. | Sources | [docs/sources.md](docs/sources.md) |
  194. | Official jaggedness | [Jev 1.13](https://docs.typesafe.ai/model-jaggedness/jev-1.13) |
  195. | Official patterns | [docs.typesafe.ai/patterns](https://docs.typesafe.ai/patterns) |
  196. | Workflow evals | [evals.typesafe.ai](https://evals.typesafe.ai/) |
  197. | Agent skills | [skills/](skills/) |
  198. ---
  199. MIT · Unofficial · [TypeSafe AI](https://typesafe.ai)

Ratings and comments

Sign in to rate this skill.

Sign in

Free to use. If it’s helped, pass it on.