const TokenCompressor = require('./src/token-compressor');
const compressor = new TokenCompressor({
  ollamaHost: 'localhost', // 默认
  ollamaPort: 11434, // 默认
  compressionModel: 'llama3.1:8b', // 默认 — 任何 Ollama 模型都有效
  maxUncompressedTurns: 10, // 保留最后 N 轮未压缩
  cacheMaxSize: 100,
  cacheTTL: 3600000 // 1 小时
});

Pre-process prompts through 3 compression layers before sending to paid APIs. Uses a free local Ollama model to do the compression work — your paid API only sees the condensed result.

Runtime Requirements

Requirement	Details
Ollama	Must be running locally (default: localhost:11434)
Local model	A small model for compression (e.g. llama3.1:8b). Configurable via compressionModel option.
Node.js	14+

Ollama is required at runtime. The compressor sends prompts to your local model — not to any external API.

What This Skill Sends to the Local Model

This skill sends the following to your local Ollama model:

Operation	System prompt	User prompt
Message compression	You are a text compression tool. Output only what is asked, nothing else.	Your message + instruction to compress
History summarization	Same	Old conversation turns + instruction to summarize

No data is sent to external APIs. All compression happens locally.

Side Effects

Type	Description
NETWORK	HTTP to localhost:11434 only — your local Ollama instance
MEMORY	Response cache stored in-memory (Map, configurable size/TTL)
DISK	None — cache is not persisted to disk

Setup

const TokenCompressor = require('./src/token-compressor');
const compressor = new TokenCompressor({
  ollamaHost: 'localhost',      // default
  ollamaPort: 11434,            // default
  compressionModel: 'llama3.1:8b',  // default — any Ollama model works
  maxUncompressedTurns: 10,     // keep last N turns verbatim
  cacheMaxSize: 100,
  cacheTTL: 3600000             // 1 hour
});

See README.md for full API documentation and usage examples.