jev-context
Codex search plugin that runs ripgrep and keeps only passages Jev judges relevant to the question.
Overview
Jev Context is a local Codex plugin for building context from a codebase. It runs ripgrep, asks Jev a yes/no relevance question for each candidate passage, and returns passages whose probability of yes exceeds a threshold (0.50 by default). Codex can recover lower-scoring passages or read surrounding code and containing functions without more Jev calls.
It targets Codex users who want smaller search results without losing relevant code. The plugin records paired filtered and unfiltered payloads so token savings can be reproduced under a named tokenizer, and defaults to ask-only mode with savings receipts on. The README describes it as an experimental MVP: relevance judgements can be wrong, candidate code is sent to TypeSafe and may incur charges, and reported savings are retrieval text only, not verified Codex usage or billing.
What the decision model decides
| Type | Question | Answers |
|---|---|---|
| Yes / no | Is this passage relevant to the search question? | Yes · No |
How it works
- Codex calls the plugin with a question and workspace
- Run ripgrep to collect candidate passages
- Ask Jev a Yes/No relevance question per passage
- Return passages above min_yes_probability; expand_results or expand_context recover more without new calls
Strengths
- Smaller retrieval payloads for Codex
- Withheld passages remain recoverable
- Reproducible token-savings records
Caveats
- Experimental MVP; relevance judgements can be wrong
- Sends code to the TypeSafe API, which may cost money
- Savings measure retrieval text only
Getting started
Install once as a local Codex plugin (instructions use Windows PowerShell and Node.js); defaults are ask-only with receipts on.At a glance
- Type
- Plugin
- Status
- prototype
- Language
- TypeScript
- Runtime
- Codex plugin (Node.js)
- Has code
- yes
- License
- MIT
Use cases and integrations
How curated lists describe it
A Codex search plugin that filters ripgrep passages through Jev before returning relevant code.
logicrw/awesome-jev-projects
Frequently asked questions
What does Jev decide?
Whether each ripgrep passage is relevant to the question Codex asked.
What if it filters something important?
Codex can call expand_results to recover lower-scoring passages without more Jev calls.
Does it cost money?
Filtered searches call the TypeSafe API and can incur charges.
Related projects
Retrieval layer for Codex that finds evidence locally, optionally ranks sanitized excerpts with Jev, and returns bounded, cited evidence.
ProjectbetaPipe grep or file lists into Jev to get every candidate scored, sorted and cut at a calibrated threshold, so agents read only what matters.
ProjectbetaTypeScriptMCP server that returns only the JS/TS code a task needs by following the TypeScript compiler graph and asking Jev about relevance.
MCP serverbetaTypeScriptnpm CLI that screens large agent tool outputs with Jev, returning compact typed decisions and IDs while originals stay recoverable.
CLIbetaTypeScriptCLI and MCP server giving coding agents Jev-powered semantic search, focused reads and list picking in about a second for about a cent.
BenchmarkbetaPythonDeterministic code-graph context for coding agents on Postgres, AGE and pgvector, with optional Jev or local re-selection of symbols.
MCP serverbetaPython