jevproxy

v1.0.0

Published

22 days ago

Sub-25ms tool calling accelerator for AI coding agents — Claude Code, Cursor, Codex, Windsurf, OpenCode. Zero reasoning tokens burned on deterministic decisions.

⚡ JevProxy — Sub-25ms Tool Calling Accelerator

Stop burning reasoning tokens on obvious decisions.

JevProxy intercepts tool-calling requests from AI coding agents (Claude Code, Cursor, Codex, Windsurf, OpenCode) and resolves deterministic ones in <25ms with zero upstream tokens — saving 80%+ on reasoning costs.

Install

npx jevproxy chat

No install needed. Runs directly with npx.

Commands

npx jevproxy chat              # Interactive coding agent session with real tools
npx jevproxy run "<prompt>"    # Single-shot prompt through JevProxy
npx jevproxy test              # Run diagnostic suite against the gateway
npx jevproxy login             # Save your JevProxy API key
npx jevproxy logs              # View intercepted decisions & token telemetry
npx jevproxy status            # Show active key & routing config
npx jevproxy claude [args]     # Run Claude Code accelerated by JevProxy
npx jevproxy codex [args]      # Run OpenAI Codex accelerated by JevProxy
npx jevproxy cursor            # One-click setup for Cursor & Windsurf

How It Works

Agent Request → JevProxy Gateway → Decision in <25ms
                                    ├─ Deterministic? → Direct tool call (0 tokens)
                                    └─ Ambiguous?     → Forward to LLM normally

When your coding agent asks "check git status" with 7 tools available, there's no reason to burn 1,400ms and $0.015 of reasoning tokens figuring out which tool to call. JevProxy resolves it in 8ms for $0.0001.

Integrate with Claude Code

npx jevproxy claude
# or set environment variables:
export ANTHROPIC_BASE_URL=https://jevproxy.com/api/v1

Integrate with Cursor / Windsurf

npx jevproxy cursor
# Prints setup instructions for IDE configuration

Integrate with Codex

npx jevproxy codex
# or set environment variables:
export OPENAI_BASE_URL=https://jevproxy.com/api/v1

API Formats Supported

OpenAI /chat/completions — GPT-4o, GPT-4, o1, etc.
Anthropic /messages — Claude 3.5, Claude 4, etc.

Both formats are auto-detected and routed correctly.

Get an API Key

Go to jevproxy.com
Sign up and grab your jev_live_... key from the dashboard
Run npx jevproxy login to save it locally

License

MIT