Six open-source projects using Jev as their decision layer

monokern @monokernOriginally posted 2026.09.25

Open original on X

JEV is tearing through GitHub right now, and the open-source community just proved what software looks like when you ban chat models from making decisions: no conversational filler, no token streaming, no 5-second latency traps. just pure 80ms execution wired into production code: jev-ultrafast (17,200+ stars): kills slow 4K screenshot loops by indexing DOM elements into compact tables. Jev triggers navigation clicks in 70ms, waking up a generative model only when text input is strictly needed. Google Flights booking runs in 7 seconds. https://github.com/browser-use/jev-ultrafast jev-trader (2,000+ stars): autonomous market maker built for Monad’s 300ms block environment. inspects order book imbalance and places post-only limit orders on Kuru DEX before the block window closes. https://github.com/jarrodwatts/jev-trader typesafe-ai/skills (2,100+ stars): the official harness for Claude Code and Codex. gives terminal agents a sub-second decision brain to inspect dangerous bash invocations and filter tool schemas before execution https://github.com/typesafe-ai/skills typesafe-mcp: universal Model Context Protocol bridge allowing external agents to offload tool routing, safety gating, and discrete choices to Jev instead of burning frontier tokens https://github.com/itsmostafa/typesafe-mcp tax-doc-classifier: production document sorter processing 261 distinct IRS tax forms with 100% corpus accuracy. classifies complete tax archives at roughly $0.001 per page with zero schema breaks https://github.com/kyotofin/tax-doc-classifier pg-jev: Postgres extension executing semantic classifications directly inside SQL queries. evaluates database rows against typed questions at native wire speed https://github.com/realZachi/pg-jev jev-curate: high-throughput data hygiene engine streaming Parquet and JSONL files at 1,500+ records per second. prunes dirty datasets for pre-training without generating text https://github.com/AkashPriyadarshii/jev-curate killmyidea: removes polite AI flattery from startup analysis. evaluates business models against market parameters and enforces a strict typed verdict: kill, fix, or ship https://github.com/monteduro/killmyidea typesafe-chess: pins Jev against a dedicated chess engine with output spaces restricted strictly to legal candidate moves. eliminates hallucinated illegal moves entirely https://github.com/TholeG/typesafe-chess jev-doom-agent: runs Chocolate Doom in WebAssembly while Jev reads game state buffers and selects tactical macros in real time on every tick https://github.com/lukaske/jev-doom-agent notice the common denominator across all ten: not one of them wastes GPU cycles synthesizing prose. they ingest raw state, evaluate typed options in 80ms, and let deterministic code execute the result. read our complete 80ms Jev engineering blueprint below before you build an eleventh ↓

This post quotes another post: x.com/monokern/status/2103153482161086753