Preview. These maps are our reading of public sources and haven't been checked with the projects' maintainers yet.
The Jev ecosystem
Projects built on or around Jev, TypeSafe AI's typed-decision model announced 2026-09-15, and the decision models and tools that speak its format.
22 projects64% active82% build on a shared standardMapped 2026-10-25
What they agree on
- How is the main thing licensed? permissive open source (MIT/Apache/BSD) (86% of 22)
- Do you need an account to use it? no account (86% of 22)
- Who is it for? developers (95% of 22)
- Which shared standard or format does it mainly build on? TypeSafe's System One API (state + typed questions, /v1/systemone) (82% of 22)
- Does it train on Jev's outputs? (TypeSafe's customer agreement forbids using output to distill or build competing models.) it doesn't train models (70% of 20)
What they argue about
Is it actively maintained?
How is it mainly sustained?
Can it run entirely on your own device or infrastructure, without the maker's servers?
Which models answer its questions? Pick the first that applies.
What is the main thing it ships?
Which question types does it handle?
Does it check whether its probabilities can be trusted? Pick the first that applies.
Where do the decisions run by default?
Can you use it without a TypeSafe account?
Are its answers typed values rather than generated text? (identity question)
Does it publish head-to-head numbers against Jev?
Projects
| Project | Status | What it is |
|---|---|---|
| AT0M (Pienomial) | Active | A private System One decision model meant to ship as a single self-contained Rust binary (CPU/Metal/CUDA); for now only a hosted evaluation… |
| Bud Decision Studio | Active | 'The LM Studio of Jev-like models': a cross-platform desktop app and local server that downloads open decision models (Kev, Laya, GLiNER2.5… |
| CUA-S1 (Cua) | Active | Cua's family of small 'System 1' models for computer-use decisions (first profile: forms, Qwen3.5-4B adapters), plus a jev-use chooser in C… |
| decide | Maintenance | Go CLI (Homebrew) for yes/no, choice and level decisions from the shell, scripts and agent skills, against TypeSafe's Jev or OpenRouter. |
| DoubtBench | Maintenance | Benchmark that scores System One models on whether their probabilities track human annotator disagreement (about 7,500 Noul/Choice/Score qu… |
| Gutsy | Maintenance | 0.8B Qwen3.5 fine-tune (GGUF, 500-775 MB) for typed decisions on an ordinary CPU via llama.cpp, with a Jev-style /v1/systemone server and b… |
| jev-align (Sutro) | Maintenance | CLI that builds calibrated 'AI Functions' on Jev: finds uncertain examples, asks you to label them, and uses GEPA prompt optimization to im… |
| Jev.directory | Maintenance | A community gallery of real systems built on Jev and the primitives (noul/choice/score) they send, with sign-in via GitHub to submit and li… |
| jeva.cpp | Maintenance | A llama.cpp fork whose llama-server adds a Jev-compatible decision API (Choice, Score, Noul from logits) for any model llama.cpp supports. |
| jevals (Openlayer) | Active | Agent evals and guardrails expressed as Jev decisions (one request per trace), with backends for Jev via TypeSafe or Vercel, OpenRouter dec… |
| JevBench (Benchmark Heaven) | Active | Independent benchmark and leaderboard for Jev-class decision models: accuracy, calibration, latency and cost, with Jev 1.13/1.14 as referen… |
| jevper | Active | Python wrapper that puts the Jev interface, and OpenAI's newer Decisions API schema, on top of any OpenAI-compatible or Anthropic model, re… |
| Jevstiller | Active | A drop-in proxy in front of Jev that learns a small local model (bge-small encoder plus logistic head) from Jev's own answers and answers l… |
| Kev | Active | Family of small Jev-like decision models (0.8B to 27B, Qwen3.5/3.8 base with adapter and pointer head) you can run or fine-tune yourself; s… |
| Laya | Active | Multilingual, non-autoregressive System 1 decision engine on ModernBERT/mmBERT (322M-421M params), trained with RL against proper scoring r… |
| onesie | Active | Unix-pipeable CLI for System One models: ask questions of piped text, gate agent shell commands by exit code, and calibrate thresholds on l… |
| OpenDecision | Active | Local typed decisions over text, documents and images using off-the-shelf zero-shot models (ModernBERT/DeBERTa NLI, SigLIP2), with a FastAP… |
| snap | Active | Rust decision engine that reads answer-letter logits from existing GGUF models in one batched pass; local, deterministic and Jev wire-compa… |
| STROT | Changed direction | Compiles a decision task once with an LLM into a frozen Python artifact; a small local model (Julia-1 by default) then makes each decision… |
| System One Adapter (official) | Maintenance | TypeSafe's drop-in replacement for the SDK's system_one call, backed by OpenAI-compatible, Anthropic or Gemini LLMs, so you can compare Jev… |
| THX-01 (HAL-X AI) | Active | 322M mmBERT-based multilingual decision model (Azerbaijani emphasis) trained with RL for calibrated decisions; Apache weights, a pip server… |
| TypeSafe Python SDK (official) | Active | TypeSafe's official client for the hosted Jev API (Python; a matching TypeScript SDK exists). Typed Noul/Choice/Score (and Box) questions o… |
Where they meet
Not mapped yet.
Follow
RSS for this community: projects added, status changes, and later polls and discussions.
Follow the whole community: OPML of every project's own release feed (where we found one) · OPML of every project's Poll Position feed.
The poll
Every project above was answered against these questions. Core questions are shared by all communities; the rest are this community's own.
- Is it actively maintained? core
active · active, but changed direction (pivot, fork, or rename) · maintenance or slowing · dormant, sunsetting, or archived - How is the main thing licensed? core
permissive open source (MIT/Apache/BSD) · copyleft or source-available with restrictions · open parts around a closed core · proprietary or closed source - How is it mainly sustained? core
individual side project, no revenue · individual with revenue (paid app, tier, ads, or donations) · community, foundation, or research · company with a paid product, tier, or ads · company or startup with a free product (no revenue yet) · big company side project - Can it run entirely on your own device or infrastructure, without the maker's servers? core
yes · yes, but it calls a third-party service you choose (your own API key or account) · partly (some features need the maker's servers) · no, it needs the maker's servers · not applicable - Do you need an account to use it? core
no account · optional account · account required - Who is it for? core
developers · non-developers · both - Is there public evidence it's built with AI coding agents? core
yes, the repo shows it (agent instruction files or agent co-authored commits) · yes, the maker says so · no public evidence - Which shared standard or format does it mainly build on? core
TypeSafe's System One API (state + typed questions, /v1/systemone) · OpenAI's Decisions API (/v1/decisions) · OpenRouter's decision-model API · none of these (its own format or interface) · not applicable - Which models answer its questions? Pick the first that applies. approach
none: it measures, compares, or catalogs Jev-style systems · only TypeSafe's Jev (direct or through a reseller) · Jev or other models, your choice · other models, not Jev (open, local, or LLMs) - What is the main thing it ships? approach
model weights · a runtime, server, proxy, or desktop app for running models · a library, CLI, or SDK for calling models · a benchmark or directory - Which question types does it handle? approach
Jev's three: noul, choice, score · Jev's three plus extra types (number, excerpt, relation, box, rank...) · a subset or a different set - Does it check whether its probabilities can be trusted? Pick the first that applies. value
yes, it publishes calibration metrics (ECE, Brier, or similar) · it calibrates or sets thresholds against your own labelled examples · no, it passes probabilities through without measuring them · probabilities are not central to it - Where do the decisions run by default? approach
on your own hardware (CPU, Mac, or GPU) · a hosted API · no default: you choose a local or hosted backend · not applicable (benchmark or directory) - Does it train on Jev's outputs? (TypeSafe's customer agreement forbids using output to distill or build competing models.) value
yes, it trains on Jev's outputs · it trains models, but not on Jev's outputs · it doesn't train models - Can you use it without a TypeSafe account? value
yes, no TypeSafe account involved · yes, through a gateway or another provider · no, a TypeSafe API key is required - Are its answers typed values rather than generated text? (identity question) value
yes, typed answers only · typed answers, but generated text is part of the pipeline · not applicable (benchmark or directory) - Does it publish head-to-head numbers against Jev? value
yes, head-to-head numbers against Jev · benchmarks, but not against Jev · no published benchmark numbers
Sources
Show HN posts 2026-07-27 to 2026-10-25 mentioning Jev, plus well-known reference projects.
46 sources across 22 projects, read on 2026-10-25. Each project page lists its own. Poll version: core-v4.