Jev in production › Agent context and memory

jev-usher (cvsgireesh)

Claude Code launcher and hooks where Jev picks Haiku, Sonnet or Opus at session start and judges which tool output enters context, keeping originals for recovery. 3.1% fewer Claude tokens with filtering across 13 synthetic tasks (author).

Open on GitHub ↗

3.1% fewermeasured against a baseline, as published by the source
Use
Agent context and memory
Industry
Developer tools
Form
Open-source tool
Stage
Beta
Plugs into
Claude Code
Listed
2026-09-20
Found via
github
Repository
cvsgireesh/jev-usher
Stars
1
Forks
0
Last push
2026-09-20
Language
TypeScript
License
MIT

The README opens with

jev-usher — model routing and context admission for Claude Code

jev-usher routes clear tasks to a suitable Claude model and filters noisy tool output, with originals kept for recovery. TypeSafe's JEV judges relevance; deterministic rules decide what to send.

Use it through Claude Code, the local comparison UI, or the TypeScript library. See measured results and tradeoffs for token use, latency, and test methodology.

Status: preview. The launcher enables routing and recoverable filtering; direct plugin and settings installations require filtering to be enabled separately. Filtering supports specific output formats and can omit useful information. Inspect decisions and compare answers in the UI before using it in ongoing work. Token reduction...

Badge

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

Listed in Jev in production

Also used for agent context and memory