agentic-rl — Benchmarks & Evaluation · Jev

A candidate-scoring head returns distributions for gates, routes and scores without decoding prose or JSON answers. The README compares a benchmark-trained AgentJev specialist with Jev 1.13.0 evaluated zero-shot; these use different training conditions.

JevBench v1 - a benchmark for Jev-class typed decision models: smart, cheap, fast, reliable, open.

A small open decision model: state + typed questions -> calibrated probabilities. A Jev / System One re-creation on Qwen3.5.

Leaderboard that puts Jev and 14 LLMs through 108 four-option multiple-choice questions about the Convex backend platform, comparing accuracy, latency, and cost, with a per-answer explorer showing Jev probabilities.

Open Jev implementation with custom finetuning

Non-autoregressive decision engine on ModernBERT (151M) with calibrated uncertainty (RLCD), TypeSafe AI Jev benchmark audit, and in-browser WebGPU playground

Instructions, tools, skills, channels and schedules live in conventional directories, making an agent project inspectable and extensible. The evaluate helper sends typed questions about supplied state through eve's model authentication, defaulting to Jev.

The native CLI supports an interactive shell, scripted requests and embedding, with skills, MCP servers and subagents. Jev receives review policy, context and the pending action, returning a clear or caution decision for the permission pipeline.

The platform supports developing and evaluating SRE agents against live system environments containing operational failure scenarios. Jev support is disabled by default; when enabled for Codex, it reviews diagnostic tests and submissions from evidence sent to TypeSafe.

No project description available.

🚀 Feather Wand - JMeter Agent for performance engineers w/ Codex and Claude Code integration

Agent orchestration that guides projects toward demonstrable goals.