prompt-chunk-trimmer

LLM Context Budget Trimmer: A zero-dependency text trimmer that counts tokens and prunes large logs, stack traces, or context blocks by removing the middle while preserving head and tail content.

⚡ Features

• Zero Runtime Dependencies: Extremely fast, lightweight, and side-effect free.
• Sliding-Window Middle Pruning: Keeps the beginning context (like system instructions or setup details) and the ending context (like the actual crash or exit logs) while trimming the verbose middle.
• Token Estimation: Fast estimations based on custom character-to-token ratios.
• Configurable Budgets: Fine-tune budget allocations for head, tail, and maximum token constraints.

📦 Installation

npm install prompt-chunk-trimmer

🚀 Usage

1. CLI Usage

Trim a large log file:

npx prompt-chunk-trimmer large-log.txt --tokens 1000

2. Programmatic API

import { trimText } from 'prompt-chunk-trimmer';

const hugeOutput = "some very long log trace...";

const result = trimText(hugeOutput, {
  maxTokens: 500,        // maximum tokens allowed (default: 2000)
  charToTokenRatio: 4,   // char count / ratio = estimated tokens (default: 4)
  headBufferPct: 0.25    // keep 25% of character budget at the top, rest at the bottom (default: 0.3)
});

console.log(result.text);
/*
some very long log

... [TRUNCATED 23412 CHARACTERS TO FIT LLM CONTEXT] ...

trace...
*/

📄 License

MIT License.