Nerfguard

Nerfguard is a local gateway that optimizes token usage for coding agents by auto-routing each request to the most appropriate model and reasoning depth. It solves the problem of wasted tokens and time on excessive intelligence for AI-assisted coding tasks. The product targets developers and engineering teams using coding agents like Claude Code and Codex, particularly those looking to reduce costs without changing their workflow. It is positioned as a free, fast, and reversible tool that works with existing providers and plans, with optional ultra routing to alternative open models for additional savings.

Key features

• Auto-routes to best model and reasoning depth
• Optimizes token usage to reduce waste
• Works with Claude Code and Codex
• Local gateway for coding-agent traffic
• Enable/disable anytime with CLI commands
• Reversible setup
• Fast classifier around 250ms
• Supports ultra routing to open models
• Provider-only routing option