01
抽出 Extract
Keeps the sentences with the highest keyword salience.
圧縮 · Compress
Awaiting input.
A note on short inputs
Compression is extractive and tuned for long, repetitive content — logs, code,
and verbose prose. On very short inputs the model-specific framing added in the
Adapt stage can outweigh the savings, so the compression figure may be
zero or negative. This is expected; the ratio turns positive as input length grows.
仕組み · How it works
Four pure stages, run in order. Each is deterministic — the same input always yields the same output.
02
統合 Synthesise
Dedupes, shortens stock phrases, abbreviates.
03
優先 Prioritise
Enforces the token budget, keeping head and tail.
04
適応 Adapt
Adds model-specific framing for Claude, GPT or Cursor.
接続 · API
The same engine is available over HTTP. Post your content to
/api/optimise and receive the compressed text plus metrics.
The command on the right reflects your current settings.
| Endpoint | Purpose |
|---|---|
POST /api/optimise | Compress one document |
POST /api/batch | Up to 50 at once |
POST /api/optimise-for-ai | Balanced, per model |
GET /api/stats | Capabilities |
GET /health | Liveness |
Example request