JEV REGISTRY • VERIFIED TOOL|DEVTOOLS & TERMINAL| 17.4MS EDGE REFLEX
Home/Hub/JevGrep
DevTools & Terminal17.4ms Reflex$0 / Apache 2.0
23,150 active downloads

JevGrep

Sub-20ms code symbol locator and regex classifier for agent workspaces

jevproxy // jevgrep (reflex kernel)
$npx jevproxy run jevgrep
JevProxy Intercept Resolved(Mode: DIRECT_REFLEX)
17.4ms|Cost: $0.0001|0 reasoning tokens burned
STDOUT • Tool Call Output:Status: 200 OK
agent.dispatch(jevgrep_search)
payload: { "query": { "type": "string", "description": "Search pattern or exact symbol" }, "path_patte...
✔ Decision short-circuited in 17.4ms without roundtrip to frontier LLM.
Upstream token bill saved: $0.0185 on this turn.
JevGrep
Compatible with Cursor, Claude Code, Windsurf, OpenCode
TRADITIONAL LLM CALL:UNOPTIMIZED
• Median Latency: 1680 ms
• Cost per Turn: $0.0185
• Mode: Full KV Cache Reload & TTFT Prefill
JEVPROXY REFLEX KERNEL:77x FASTER
• Median Latency: 17.4 ms
• Cost per Turn: $0.0001
• Accuracy: 98.9% deterministic

1-Click CLI Execution

Run this tool accelerated through the JevProxy gateway without manual wiring:

npx jevproxy run jevgrep

OpenAI / Anthropic Tool Schema

JSON Specification

Paste this schema into your agent tools definition or Cursor extensions:

{
  "type": "function",
  "function": {
    "name": "jevgrep_search",
    "description": "Fast-path symbol search and context bounding across workspace files",
    "parameters": {
      "type": "object",
      "properties": {
        "query": {
          "type": "string",
          "description": "Search pattern or exact symbol"
        },
        "path_pattern": {
          "type": "string",
          "description": "Glob filter for target paths"
        },
        "max_depth": {
          "type": "number",
          "description": "Directory recursion limit"
        }
      },
      "required": [
        "query"
      ]
    }
  }
}

Technical Architecture & Usage

The File Search Latency Problem

Standard coding agents spend up to 45% of their turn time executing ripgrep commands and waiting for frontier models to parse the terminal output. A single search query often inflates context by 6,000 tokens.

The JevGrep Solution

JevGrep executes local ripgrep passes and passes candidate matches through an 18ms JEV classification head. It filters out irrelevant dependencies and node_modules noise before transmitting to the primary agent model.

Accelerate JevGrep with JevProxy

Get 5,000,000 free decision tokens. Eliminate the 3-second tool freeze in Cursor and Claude Code in under 60 seconds.