TokenLens: A Token Waste Profiler

Token cost is a design problem, not a billing problem. Most LLM cost overruns come from architectural waste, not model pricing.

This tool is a token waste profiler that helps you understand where your tokens are going and which ones are useless. It's like valgrind for LLMs.

The Problem

Most LLM systems waste 30-60% of their tokens on:

• Repeated instructions
• Bloated context
• Unused history
• Overlong outputs

Teams have no visibility into this waste. That's where TokenLens comes in.

What TokenLens Is Not

• A token counter
• A billing dashboard
• A UI with charts
• A cost calculator

What TokenLens Is

A token waste profiler that:

• Wraps LLM calls
• Tracks token-in vs value-out
• Identifies repeated context, redundant instructions, and oververbose responses
• Produces actionable recommendations

Usage

python cli.py --run examples/sample_calls.json

This will output a JSON report with a breakdown of token waste and recommendations for optimization.