What it is
A Pi extension for developers running long sessions. Instead of asking an LLM to summarize old turns, it asks Jev to score each tool call and drops only what Jev says is no longer needed.
How it uses Jev
Per non-pinned tool call, Jev answers two nouls: call_tN (does knowing the call still matter) and result_tN (must its full output stay verbatim). Keep probabilities are compared to keepThreshold to keep, truncate, or drop.
Primitives:noul
Technique worth stealing
Selective, verbatim dropping driven by per-call model scores instead of lossy LLM summaries.
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-centricyes0.92
- Shows a System One patternno0.37
- Handles uncertaintyno0.05
- Measuredyes0.91
- Runnableyes0.74
- Worth recommendingunclear0.45
- Model replicano0.04
- Problem scopescore on a 0–2 scale1.04
- About Jevyes0.98
Signals by Jev jev-1.13.0, card written by DeepSeek V4.1 Flash from the README on 20 Sept 2026.