Open source around Jev

Curated list of public projects, integrations and discussions built on Jev — including MCP servers that expose classify, score, check, match and screen as tools for any agent, and a ZIO/Scala client with a typed DSL.

Asks whether something Jev-like can run on a 3090 at home. Reproduces the interface pattern with open models rather than TypeSafe's undisclosed model or training. Needs Python 3.10+, CUDA, and a GPU that fits a 4B BF16 model.

Benchmarks Gemma 3 4B zero-shot against TypeSafe's published Jev numbers and implements a /v1/systemone endpoint matching the documented request/response shape. Finds Gemma over-confident on judgement calls — a useful read on the zero-shot ceiling.

An independently trained starter model with the same input/output shape. Its option-attention head can also score controller buttons from image patches.

Jev versus Claude Haiku 4.5 on 2,000 phishing emails — accuracy, calibration, latency and cost, fully reproducible. Found Jev's raw verdict statistically worse, but a logistic regression over five signal questions from the same call reaching 95.1% accuracy.

Reproducible benchmark for Jev on agent tool-call risk classification: accuracy, latency, and whether the confidence score is worth routing on. Comparison columns are empty pending provider API keys.

A browser agent with a dynamic, indexed action space: Jev picks the operation and the target element, and a small LLM writes text only for TYPE_TEXT operations.

A personal-assistant agent with 100 mocked tools that compares how many steps are needed when the LLM picks the tool itself versus when Jev picks it. Implements confidence-gated routing with a configurable threshold.