JEV Plugins & Tools Directory
The fast-growing ecosystem around TypeSafe's JEV ("System One") models — decision models that return typed outputs instead of text. Ranked by GitHub stars, auto-refreshed. Also see our JEV tutorial.
What is a JEV (System One model)?
A JEV is a "System One" model released by TypeSafe AI on 15 September 2026. Unlike a chatbot or LLM, it never writes sentences — it returns a typed decision(a label, a score, a route, a yes/no) with a calibrated confidence number. That single design choice makes it 40-400x cheaper and ~150ms fast, so it fits inside agent loops where a 3-second text answer is too slow and too expensive.
The name is a deliberate nod to Daniel Kahneman's Thinking, Fast and Slow: System One is fast instinctive judgment, System Two is slow deliberate reasoning. JEV handles the fast half — the constant routing, classification and scoring that agents do thousands of times a day — while a larger LLM handles the reasoning that actually needs words.
Choosing between them? See Laya vs Jev vs Kev →
Top JEV ecosystem projects
| # | Project | Type | Stars |
|---|---|---|---|
| 1 | NandhaKishorM/laya Non-autoregressive System 1 decision engine — the open-weight alternative to TypeSafe Jev. Typed choice, score and yes/no decisions you can run locally. | Model | ⭐ 19.4k |
| 2 | browser-use/jev-ultrafast Fastest and cheapest web agent built on JEV-style typed decisions instead of token generation. | Agent | ⭐ 18.8k |
| 3 | tamaratran/fast-jev-compaction Claude Code plugin that replaces the compaction summary with Jev decisions — every pruned turn becomes a typed, replayable judgment. | Plugin | ⭐ 6.3k |
| 4 | jaredpalmer/kev Tiny Jev-like family of decision models built on top of Qwen — open-weight System One alternatives for local routing and classification. | Model | ⭐ 4.2k |
| 5 | AbdelStark/awesome-typesafe Curated list of official TypeSafe resources, System One papers and community projects. | List | ⭐ 380 |
| 6 | NiazMorshed2007/jev-review Local-first MCP plugin: JEV provides structured code-review feedback to Claude Code, Codex, Cursor and OpenCode. | MCP | ⭐ 280 |
| 7 | fatwang2/awesome-jev Source-backed JEV project collection: classification experiments, A/B input comparison, conversation routing utilities. | List | ⭐ 450 |
Why the JEV ecosystem is exploding
- Cost: System One decision models run 40-400x cheaper than LLM calls for routing, classification and scoring tasks.
- Speed: ~150ms typed decisions fit real-time agent loops where 2-10s text generation does not.
- Integrations: Vercel AI Gateway added JEV as a provider; Claude Code plugins (fast-jev-compaction) replace fuzzy summarization with replayable decisions.
- Open alternatives: Community models like kev (Qwen-based) bring System One-style decisions to local hardware.
Star counts are point-in-time snapshots from the GitHub API and change fast in this ecosystem. Also browse our DeepSeek Harness directory and M3E Canvas tutorial.