kydlikebtc/awesome-jev

Overview

awesome‑jev is a curated catalogue of public examples that use Jev, a typed‑decision‑model service from TypeSafe AI. Jev does not generate free‑form text; instead you send it a state and a set of typed questions and it returns structured answers (choice, score, or “noul”) together with a calibrated confidence score. The catalogue indexes real‑world code‑snippets, tutorials, benchmarks and measurements that show how developers have integrated Jev into agents, tools, and pipelines.

The repository itself does not contain an SDK, a model implementation, or any runtime code for Jev. Its purpose is to record what has been tried, with links to the original source files, brief notes on any caveats, and – where available – independent measurement results (including negative findings). All entries are stored in catalog.json and rendered on a searchable web site.

---

What the project provides

Component	What you get	How it’s useful
Catalog of examples	Over 1,200 public resources (code, tutorials, benchmarks) organized by decision pattern (e.g., tool selection, safety gating, content scoring).	Quickly find a concrete snippet that matches the decision your agent needs to make.
Pattern index	18 decision patterns, each with a count of examples (e.g., 230 for Tool selection, 166 for Content scoring).	Helps you see where the community has the most experience and where gaps remain.
Primitive guide	Visual description of Jev’s three primitives – choice, score, noul – and a decision‑list picker that tells you which primitive to use for a given problem.	Reduces the learning curve for new users of Jev.
Measured‑vs‑claimed section	Independent benchmark reports (both favourable and unfavourable) with author‑stated conclusions.	Lets you assess real‑world performance before adopting a pattern.
Machine‑readable data	catalog.json, picker.json, and other CSV/JSON assets that can be consumed by scripts or the hosted site.	Enables automated analysis or integration into your own tooling.
Contribution guide & licence	Instructions for adding new entries and a permissive licence.	Encourages community growth and keeps the catalogue up‑to‑date.

---

Who might benefit?

Audience	Why it matters
AI agents / tool‑calling developers	Need concrete, vetted examples of how to phrase Jev questions and interpret its typed answers.
Researchers evaluating decision‑model services	Can compare Jev‑based approaches against alternatives using the independent measurement reports.
Product teams considering Jev	Get a realistic sense of integration effort, cost, and known limitations (e.g., no local weights, 64 k token context).
Educators & tutorial authors	Have a ready‑made list of real‑world code to illustrate typed decision‑making.

---

Limitations & what’s not covered

• The catalogue does not run any code; links point to source files that were read, not executed.
• Jev’s model weights are proprietary – you cannot run the model locally; all calls go to the TypeSafe API.
• Coverage counts reflect saved records, not current link availability or CI status.
• The repository does not provide a unified SDK; you still need to use the official TypeSafe client libraries.

---

Quick start to using the catalogue

1. Visit the web UI (link at the top of the README) and pick a decision pattern that matches your use‑case.
2. Browse the list; each entry shows language, primitive(s) used, and a star‑band indicating repository popularity.
3. Click the call‑site link to see the exact request shape that was cited.
4. If you need confidence about performance, filter to the Measured collection to see independent benchmarks.
5. Add your own example by following the contribution guide – the site will automatically rebuild the README.

---

Bottom line

awesome‑jev is a well‑maintained, community‑driven index of how developers are applying TypeSafe’s Jev decision model. It does not replace the official SDK or documentation, but it dramatically speeds up discovery of real‑world patterns, highlights known pitfalls, and aggregates independent performance evidence – all valuable for anyone building AI agents that need reliable, typed decisions.