SuperCompress documentation

v2 is a 400M-parameter engine that keeps the context that answers the query and cuts the rest. Open source (MIT). API, and an agent plugin for 40+ harnesses.

⚡

Quickstart

Install, get a key, compress your first prompt — in under 5 minutes.
📖

API Reference

Python API, HTTP endpoints, response fields, error codes, and client usage.
🔌

Integrations

Wire SuperCompress into OpenAI, LangChain, Express.js, Vercel AI SDK, and more.
🚀

Usage

Call the hosted API or pip install locally — no servers to manage.
🧠

Architecture

How the learned policy works: features, training, inference, and compiler mode.
⚙️

Compress engine

Layman + technical notes for every function in compress-engine.js.
🌱

Environment & CO₂

Methodology for estimating energy and carbon savings from token reduction.
🤖

Coding Agents

Install the proxy — one command to compress context for Cursor, Windsurf, Claude Code, and more.

What is SuperCompress?

SuperCompress v2 is a 400M-parameter, query-aware context engine — Neural Keep is live on the hosted API. It looks at the full context, keeps what is relevant to answering the query, and cuts the rest. On the coding-agent benchmark it cut context 64.1% on average, held the required evidence in 24/24 cases, and took one dump from 16,647 tokens to 5,148. One case reached 96.6% reduction at ≥99% evidence retention.

It is strongest on coding-agent dumps: tool output, logs, diffs, and stack traces. Unlike truncation (which blindly drops the tail) or summarization (which rewrites IDs and errors), SuperCompress keeps the original lines that answer the current question. The query itself is never compressed.

Key facts

Metric Value
Engine~400M Neural Keep (hosted, live) · query-aware
Coding-agent benchmark64.1% mean cut · 24/24 evidence held · 16,647 → 5,148 tokens
Best case on that suite96.6% reduction at ≥99% evidence retention
AccessAPI and an agent plugin for 40+ harnesses
Hosted priceLaunch promo (live): 5M free tokens/month, then $0.10 per 1M tokens
LicenseMIT — free for commercial use

When to use it

SuperCompress is designed for production AI features where context grows over time and token costs scale linearly with usage:

  • Chatbots & conversational AI — long chat histories where each turn appends to the context window
  • RAG pipelines — retrieved chunks that overshoot the context budget
  • AI search — indexing results, snippets, and document extracts sent with every query
  • Support copilots — customer transcripts, ticket history, knowledge base articles
  • Agent loops — tool call traces, intermediate outputs, log entries accumulating across turns
  • Memory systems — compressed episodic context for persistent agent memory
Not sure if it fits? Try the interactive playground — paste your actual context and see the compression results in real time. No signup needed.

How it works (briefly)

In compiler mode (the default), SuperCompress:

  1. Segments the input context into semantic blocks (headings, code blocks, log entries, prose sections)
  2. Scores each block against the user's query using a learned policy
  3. Removes duplicate content, noise, boilerplate, and low-value blocks
  4. Preserves context dependencies (nearby headings, imports, trace context)
  5. Reports tokens saved, important context kept %, and compression risk level

The full pipeline takes ~60ms on a single CPU core — no GPU, no model download, no extra LLM calls.

SuperCompress · GitHub · Home · Dashboard · MIT License