Jev in production › Agent context and memory

JIT-JEV Context OS

Coding-agent context runtime where a Jev gate shadow-evaluates context facts, proposes evictions and prefetches memory, failing open to heuristics. On fixture tasks, Jev runs showed -31.3% turns per task vs a full-context baseline, with fewer solved (author).

Open on GitHub ↗

-31.3%measured against a baseline, as published by the source
Use
Agent context and memory
Industry
Developer tools
Form
Open-source tool
Stage
In production
Plugs into
Agent Zero
Listed
2026-09-24
Found via
github
Repository
wojciechwiesner/jit-context
Stars
9
Forks
1
Last push
2026-11-14
Language
Python
License
MIT

The README opens with

JIT-JEV Context OS: Epistemic Runtime & System 1 Gate for AI Agents

A deterministic context runtime for autonomous coding agents with tool access. It cuts agent turns and wall time on discovery-heavy coding tasks. It is not a universal token saver: on short, single-file tasks it adds a small capsule overhead.

Every number below links to a committed artifact. Single runs are marked N=1; treat them as case studies, not population estimates.

Run Setup Turns Wall time Prompt tokens Pass :--- :--- :---: :---: :---: :---: Synthapse feature, 45 modules gemini-3.8-flash, N=1 171 → 66 (−61%) 568 s → 238 s (2.38x) not recorded both 10-task SWE-style fixtures 10 tasks × 3 modes, N=1 each 69 → 49 (−29%), fewer in 9/10 208 s → 132 s (−36%), faster in 7/10...

Badge

For the project's own README, linking back here:

Listed in Jev in production

Also used for agent context and memory