What is the catch with the Jev mode?

It sends request content and candidates to TypeSafe, streaming stays exact-only, oversized state skips Jev, and a replay returns a stored response — not proof an external action ran.

How do I wire it into Claude Code?

One install command for your agent, then point OPENAI_BASE_URL at the local server. Every response carries HIT, MISS, or BYPASS headers plus a jev-versus-exact match header.

Will it cache my API keys or private prompts?

It should never see them: the README says to keep keys, passwords, and credentials out of cached prompts, and the TypeSafe key is kept separate — never put into cached requests or forwarded upstream.

What does the CLI give me, and which agents does it set up?

The npx CLI has start, install, init, stats, clear, env and doctor commands, and one-command setup for Codex, Claude Code, Cursor, OpenClaw and Hermes. There is a zero-dependency Node proxy for npm and a SQLite-backed Python server sharing the same OpenAI-compatible routes.

How do I tell whether a request hit the cache?

Every response carries X-Computer-Use-Cache: HIT, MISS or BYPASS, plus X-Computer-Use-Cache-Store for stored entries. A semantic Jev hit is marked X-Computer-Use-Cache-Match: jev, while an ordinary JSON exact hit is marked exact.

What are the Jev semantic-reuse defaults and limits?

JEV_MODEL defaults to jev-latest and JEV_MIN_CONFIDENCE to 0.95, with at most 8 candidates, a 24,000-character input cap and a 10-second timeout. Streaming requests use exact caching only, oversized comparison state skips Jev, and the thresholds are operating defaults rather than validated accuracy guarantees.

CommunityJev Demos & Experiments

Computer Use Cache

Q: My workflows repeat the same prompts. What gets reused?

Exact repeats serve locally. With a TypeSafe key, Jev additionally judges whether a cached response is semantically reusable for the current request — or returns none and falls through to the upstream provider.

Drop-in OpenAI-compatible cache proxy for repeatable agent workflows: exact hits stay local, and with a TypeSafe key Jev judges whether a cached response is semantically reusable for the current request.

See highlights

Open github.com

Stars: 45
Language: JavaScript
License: MIT
Published by: Community
Author: rohanarun
Last verified: Oct 2026

rohanarun/computer-use-cache

User repository on GitHub

View on GitHub

Drop-in OpenAI-compatible cache that replays repeated computer-use and agent workflows at near-zero cost.

Forks: 11
Last pushed: Updated Sep 2026

Repo stats from the GitHub API, cached Oct 2026.Homepage

About this project

Stop paying twice for the same computer-use task. Replay repeated browser, coding, and tool workflows at near-zero cost through a drop-in OpenAI-compatible API. Benchmark preview: 256/256 cache hits, 100% computer-use cost saved, 60%+ faster runtime. Try the live Super API dashboard.

Quoted from the project's README. Read the full README.

Tagscommunity jevbest catalog-import

Works forJev for Classification Jev for Evaluation

Used inJev for Software & DevTools

Languages: JavaScript

Highlights

Opt-in Jev semantic reuse compares the current request against cached ones and picks an unchanged response or none; failures fall through upstream.
Semantic hits are marked X-Computer-Use-Cache-Match: jev; jev-latest at 0.95 minimum confidence is the default, both configurable.
Zero-dependency Node proxy or SQLite-backed Python server, with one-command setup for Codex, Claude Code, Cursor, OpenClaw, and Hermes.
Opting in sends request content and candidates to TypeSafe; streaming requests use exact caching only. MIT licensed.

Watch out

Replay returns a stored response, not proof an external action ran; evaluate thresholds on your own workflows first. Copy reviewed against the repo README on 2026-10-04, not from running it.

Questions about this entry

Short answers drawn from the linked project or page.

More Jev resources like this

sift

Chrome extension that labels each X post with Substance, Humor, Chit-chat, Promo or Junk plus AI-written and off-topic flags, collapsing hidden posts to one line.