Why trust a model adjudicator over cosine similarity caching?

Cosine caches match vector closeness while this asks a System One adjudicator for a calibrated same-intent verdict. The repo reports zero false positives on a live 100-pair run against a Jaccard baseline, but treats that as precision-first practice rather than a guarantee.

Do I need an OpenRouter key to run this locally?

Cloud Jev via OpenRouter is the default and needs OPENROUTER_API_KEY. Local Kev, Laya, Laya MLX, or any System One server needs no OpenRouter key for admits, though upstream chat still needs a real key or MOCK_UPSTREAM=1.

What gets cached when my agent uses tools or streaming?

Final assistant text is stored and a later stream of that answer can hit, while any response containing tool_calls is returned but never stored. Streams, tool schemas, and history bypass when TURN_CACHE is off.

How do I prove MISS-then-HIT wiring in my own app?

Point the client baseURL at http://127.0.0.1:8080/v1, run start --demo or the MISS-then-HIT paraphrase check, and inspect X-Jevcache headers or /stats. Without a judge it runs exact-only where only identical prompts hit.

Which adjudicators can decide a semantic hit?

Cloud Jev through OpenRouter by default, or a local Kev, Laya, Laya MLX, or any System One server selected with the ADJUDICATOR variable, with no OpenRouter key needed for admits.

What does jevcache refuse to cache?

Any response containing tool_calls is returned and never stored; it stores final assistant text only, and a later identical stream of that answer can HIT while the first stream is still a MISS.

When should I not put jevcache in front of an app?

For personalised or secret questions such as account balances, agents that must re-run every tool step, and multi-tenant cloud use without a tenant header, since a job id is not a tenant boundary.

CommunityJev Agents & MCP

Jevcache — kushals256

Local OpenAI-compatible proxy that skips repeat model calls on same-intent paraphrases and retried agent turns, judged by Jev or a local System One model.

See highlights

Open github.com

Stars: 12
Language: TypeScript
License: MIT
Published by: Community
Author: kushals256
Last verified: Oct 2026

kushals256/jevcache

User repository on GitHub

View on GitHub

MorrowCache — skip the chat call when the question is the same. OpenAI-compatible proxy. npm: @kushalicious/jevcache

Forks: 0
Last pushed: Updated Oct 2026

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

This is one of several Jev listings for the same project. The canonical entry is Jevcache — hyperspaceai.

About this project

MorrowCache — routers pick a model. This proxy decides whether to call one. npm package and CLI stay @kushalicious/jevcache / jevcache (download history unchanged). Site: morrowcache.vercel.app.

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

Tagscommunity awesomejev catalog-import

Works forJev for Classification Jev for Routing

Used inJev for Software & DevTools

Languages: TypeScript

Highlights

Drop-in /v1/chat/completions proxy; change baseURL with no SDK changes required.
Same-intent paraphrase admits plus agent-turn cache keyed on prior plus latest ask.
Fails open on adjudicator errors and expires reuse through freshness checks.
Node built-in SQLite, /stats and /healthz endpoints, and npx start --demo flow.

Watch out

Same intent is time-bounded and freshness can refuse reuse; personalized content, tool-call replays, and multi-tenant use without a tenant header are out of scope. Copy reviewed against the repository 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

jev-code

TypeScript SDK and CLI coding agent where Jev only picks options while your code runs effects, building programs as syntax trees.

Jev Agents & MCP#community#awesomejev#catalog-import

Community

rhighs/jev-codeTypeScript

pi-jev-router

Pi extension that classifies each task with Jev over OpenRouter and routes it to a Pareto-optimal model balancing measured quality, cost, and latency under a role policy.

Jev Agents & MCP#community#jevbest#catalog-import

Community

philippdubach/pi-jev-routerTypeScript

jev-router

Node router that sends each query to the cheapest capable LLM tier by asking Jev to classify demand against your tier descriptions, with no LLM call spent on routing.

Jev Agents & MCP#community#jevbest#catalog-import

Community

rajdhakad9826/jev-routerTypeScript

jev-guard

Claude Code plugin that scores pending tool calls and finished answers with Jev noul probabilities, denying the call with the correction attached when a rule threshold is met.

Jev Agents & MCP#community#awesomejev#catalog-import

Community

muratcakmak/jev-guardTypeScript

pijev

Terminal coding agent on Pi where Jev shortlists skills, ranks retrieved excerpts, and triages tool failures while the coding model handles reasoning and edits.

Jev Agents & MCP#community#awesomejev#catalog-import

Community

tonyzdev/pijevTypeScript

jev-model-router

Claude Code routing mod that asks Jev which model tier and effort each turn needs, then directs that turn to haiku, sonnet, opus, or fable with visible route lines.

Jev Agents & MCP#community#awesomejev#catalog-import

Community

satviksinha/jev-model-routerTypeScript

Back to all resources

From the Jev community

Posts from builders shipping with Jev right now.

Follow @typesafeai