jev (BorisLeMeec)
Go-based Claude Code plugin where Jev reads code so the agent doesn't: semantic find, bounded questions and large-read takeovers.
Overview
jev by BorisLeMeec is a Claude Code plugin, written in Go, that answers questions about a codebase without reading files into the agent's context. Its hook checks every turn but only takes over searches and large reads; updates, Bash calls and files under 400 lines pass straight through. Commands such as jev find locate code by description, for example finding a register flow in a repo where the word register never appears because the flow is called signup and the UI is in French.
The motivation is measured: across nine real Claude Code sessions, cache reads of re-sent context totalled about 422 million tokens while all tool results combined were about 196 thousand, or 0.04% of the total, with roughly 184 thousand tokens re-sent per turn. So the lever is keeping bytes out of context in the first place, since a 3k-token file read with 100 turns remaining costs around 300k tokens. The plugin uses Jev at $0.042 per million tokens, installs via the Claude Code plugin marketplace, needs Go 1.22+ and a key, includes a probe command to verify the API contract, and is MIT licensed.
What the decision model decides
| Type | Question | Answers |
|---|---|---|
| Yes / no | Does this file or region match the described code (e.g. the register flow)?(inferred) | yes · no |
| Score | How relevant is this location to the question?(inferred) | — |
How it works
- Agent issues a search or large read
- Hook intercepts and runs jev find or a bounded question
- Jev scans files and scores matches
- Only the matching locations or answer enter the agent's context
Strengths
- Keeps large files out of context, where most token cost accrues
- Finds code by meaning even when names differ
- Small and medium operations pass through untouched
Caveats
- Requires Go to build and a TypeSafe key
- Results quality depends on Jev's per-file judgments
Getting started
`claude plugin marketplace add BorisLeMeec/jev`, `claude plugin install jev@jev`, then `export TYPE_SAFE_AI_KEY=...`.At a glance
- Type
- Plugin
- Status
- beta
- Language
- Go
- Runtime
- Claude Code plugin, CLI
- Has code
- yes
- License
- MIT
Use cases and integrations
How curated lists describe it
A Go Claude Code plugin using Jev to find files, answer bounded questions across code and handle large reads.
logicrw/awesome-jev-projectsA claude code plugin for jev.
hellogumbo/awesome-jev
Frequently asked questions
What does the plugin take over?
Searches and large file reads; small reads and other tools pass through.
Why does this save tokens?
Context is re-sent every turn, so avoiding large reads saves far more than shrinking tool output.
Can I use it outside Claude Code?
Yes, `make install` puts the jev CLI in ~/.local/bin.
Related projects
MCP server that returns only the JS/TS code a task needs by following the TypeScript compiler graph and asking Jev about relevance.
MCP serverbetaTypeScriptRetrieval layer for Codex that finds evidence locally, optionally ranks sanitized excerpts with Jev, and returns bounded, cited evidence.
ProjectbetaDeepSeek Harness jev_search tool that runs ripgrep, scores every hit line with Jev, and returns only the top rows above a cut.
PluginbetaJavaScriptPipe grep or file lists into Jev to get every candidate scored, sorted and cut at a calibrated threshold, so agents read only what matters.
ProjectbetaTypeScriptCodex search plugin that runs ripgrep and keeps only passages Jev judges relevant to the question.
PluginprototypeTypeScript- winnow★ 110
Calibrated context sieve for Claude Code: Jev judges each block of large tool output and hides confident-no blocks behind recallable stubs.
Projectbeta