GitHub

Ask typed questions about text or JSON and inspect model-derived probability distributions in a browser. One container runs the model, the original Decision Lab explorer, and a TypeSafe-shaped API.

Run Jev-style typed decisions locally on your Mac with low RAM usage and fast responses.

Run SemIf (Jev-style semantic-if decisions) on a CPU — no GPU. Reads typed option probabilities straight from an open model in one forward pass, plus a web UI.

Super fast Jev like server, model agnostic, working with any OpenAI compatible endpoint.

open-jev is a browser-focused TypeScript library for typed decisions: one piece of text (the state) plus any number of typed questions go in, and one forward pass returns a calibrated probability distribution per question. Nothing is generated, so an answer is always one of the options you provided.

Supersimple way to serve LLMs as a Jev-like endpoint.

Jev-style structured decisions on DiffusionGemma: the example server from vLLM PR 57250.

Open reproduction of TypeSafe Jev: a 150M typed decision engine (noul/choice/score in one non-autoregressive pass, calibrated confidence). 0.697 vs Jev's 0.727, 2.5x better calibrated, 4x faster, free. Trains on a Colab T4 in 30 min.

Train small, insanely fast local classifiers from Jev-compatible examples. Run locally in your browser or Node.js.

Turn any LLM into a Jev-style decision model: typed decisions, real probabilities, no training. (continue updating).

Open probability decisions with Qwen3.5-2B, Qwen3.5-9B and Qwen3.8-27B.

Run Laya, the open-source Jev-compatible System-1 decision model, from Node.js / TypeScript via ONNX Runtime.

Ten runnable agent-decision examples covering memory, retries, context selection, and handoffs, with offline fixtures and optional live Jev calls.

Fifty runnable visual demos comparing Jev typed decisions with an OpenAI structured-output baseline, with fixtures and explicit methodology limits.

Editable Game of Life ruleset bench where Jev evaluates legal cells and chooses each intervention while code advances and verifies the board.

Commit-reveal rock-paper-scissors game where Jev scores player behavior, predicts the next move, and chooses its own move before the player acts.

SRT translation app whose Jev Noul checks flag subtitle pairs that need human review without rewriting them.

LangChain and Flask agent where Jev chooses among knowledge, calculator, weather, and chat routes, with a local-rule fallback.

TypeScript library that maps Jev Choice, Score, and Noul questions onto classification, ranking, extraction, verification, and array workflows.

Python library that maps Jev Choice, Score, and Noul questions onto classification, ranking, extraction, verification, and DataFrame workflows.

Local 596M typed-decision model with its own weights, Choice, Score, and Noul support, and a Jev-compatible System One endpoint.

Independent DiffusionGemma and SGLang typed-decision server for text and images, with a drawing playground and public evaluation artifacts.

MuJoCo xArm7 study where Jev chooses movement directions and gripper actions from physical feedback.

Classify an artist's discography by theme, mood, and lyrical complexity with Jev (TypeSafe AI), and view it as a colorful terminal dashboard

A janitor for markdown vaults powered by TypeSafe Jev: Jev votes on each note, your code files it, you review the low-confidence pile.

EmbodiedJev: MuJoCo robot decision workbench with MiniCPM5-2B, Jev and compatible model APIs

Chrome extension for systematic-review screening where Jev returns per-criterion match probabilities and an inclusion judgment.

Security-alert triage filter where Jev or a local model returns typed dispositions while policy routing remains in code.

Metis: automatic GitHub issue triage powered by TypeSafe AI Jev. A reusable GitHub Action.

Local-first document evaluation workspaces powered by Jev

Document OCR router where Jev judges page by page whether the existing text layer is usable or OCR is required.

Call-coaching application where Jev evaluates conversation context to surface structured coaching decisions.

Ableton Live controller where Jev maps a short natural-language request to a bounded music-production action.

Android conversation copilot where Jev interprets chat context and helps rank candidate replies before the user inserts one.

Claude Code and Codex browser skill that asks Jev to choose the next browser action.

Browser automation MCP server where Jev chooses the next bounded browser action for a coding agent.

Browser-agent integration where Jev selects typed actions for research and classification workflows.

Desktop automation runtime with a Jev skill that chooses UI operations and targets from accessibility-tree candidates.

Agent-computer platform with a runnable pull-request review lab where Jev judges whether findings are introduced, supported, and actionable.

Computer-use platform with a Jev recipe where the model selects one bounded action candidate and the client validates it before execution.