context-compactor

Cloudflare Worker that compresses long text to fit memory limits — no external API calls required.

What This Gives You

• Extractive summarization — scores sentences by word frequency, keeps the most informative ones
• Sliding window compression — chunks text into summary segments with configurable window size
• Keyphrase extraction — pulls top bigrams as a compact representation
• Zero dependencies — runs entirely on-Worker, no LLM or third-party API
• KV-cached results — SHA-256 keyed cache with 24-hour TTL

Quick Start

# Deploy to Cloudflare Workers
wrangler deploy

# Compress text via API
curl -X POST https://context-compactor.<your-subdomain>.workers.dev/api/compact \
  -H "Content-Type: application/json" \
  -d '{"text": "Your long text here...", "strategy": "extractive", "ratio": 0.3}'

Strategies

Strategy	Description	Best for
extractive	Top-N sentences by TF scoring	General compression
sliding	Windowed chunk summaries	Very long documents
keyphrase	Top bigrams as keyphrases	Keyword extraction

Response

{
  "compacted": "Compressed text...",
  "strategy": "extractive",
  "inputLength": 5000,
  "outputLength": 1500,
  "compressionRatio": 0.3
}

How It Fits

A Cocapn Fleet vessel. Part of the SuperInstance ecosystem for distributed AI operations.

Related repos:

• cocapn-shells — fleet shell infrastructure
• context-compactor-v2 — next-gen compactor
• context-lattice — context window management

License

Apache 2.0