jevbooks

← All projects

jevloop

parkavenue9639/jevloop

Jev-driven agent runtime with paired LLM-only comparison.

What it is

JevLoop is an open-source Python agent runtime for tool-using agents. Jev makes fast typed tool decisions, an LLM authors text and reviews low-confidence choices, and deterministic code owns execution, safety, durability, and state.

How it uses Jev

Jev decides operations and compatible targets via typed probability distributions compiled from mounted ToolSpec declarations. Code consumes only the target head matching the selected operation. Low-confidence decisions escalate to an LLM review; Jev may also bind observation references or choose LLM_PARAMETERS.

Primitives:choice

Technique worth stealing

Compile tool specs into one Jev request with an operation question plus speculative target questions, consuming only the compatible target head.

Try it

cd backend && uv sync && uv run jevloop smoke

View on GitHub

judged by Jevjev-1.13.0

Evidence

Each line is one question put to Jev about the README. ≥ 0.60 reads as yes, ≤ 0.40 as no; in between Jev is not making a call.

  • Jev-centricyes0.96
  • Shows a System One patternyes0.94
  • Handles uncertaintyyes0.80
  • Measuredno0.05
  • Runnableyes0.84
  • Worth recommendingyes0.62
  • Model replicano0.06
  • Problem scopescore on a 0–2 scale1.66
  • About Jevyes0.98

Signals by Jev jev-1.13.0, card written by DeepSeek V4.1 Flash from the README on 23 Sept 2026.