BEAVER OpenClaw Plugin

Wraps BEAVER long-text compression as an OpenClaw plugin. Type /beaver <your question> to automatically compress conversation context and get an answer.

Architecture

User: /beaver What are the main contributions of this paper?
  → OpenClaw collects long text from conversation
  → Calls BEAVER API to compress
  → Compressed text + question → LLM answers

Quick Start

1. Start the BEAVER API Server

# Install PyTorch (see https://pytorch.org/get-started/locally/ for your platform)
pip install torch --index-url https://download.pytorch.org/whl/cpu
pip install transformers fastapi uvicorn

python beaver_server.py --model_path Qwen/Qwen3-0.6B --port 8765

2. Install the Plugin

# From npm (recommended)
openclaw plugins install openclaw-beaver-plugin

# Or from source
git clone https://github.com/JusperLee/BEAVER.git
openclaw plugins install ./BEAVER/openclaw-plugin

3. Usage

In an OpenClaw conversation:

# Paste your long text first, then:
/beaver What are the main contributions of this paper?
/beaver Summarize the key findings
/beaver What is the core logic of this code?

Capabilities

| Method | Description | |--------|-------------| | /beaver <question> | Slash command — auto-collects conversation context, compresses, and asks | | beaver_compress tool | Pure compression — returns compressed text + stats | | beaver_ask tool | Compress + format prompt for LLM in one step |

Configuration (Optional)

{
  "plugins": {
    "beaver-compression": {
      "serverUrl": "http://127.0.0.1:8765",
      "defaultPageSize": 64,
      "defaultTopK": 22
    }
  }
}

API Reference

POST /compress

Request:

{
  "context": "Your long text...",
  "query": "Your question",
  "page_size": 64,
  "anchor_pages": 4,
  "flow_window": 4,
  "flash_top_k": 22,
  "semantic_weight": 0.7,
  "lexical_weight": 0.3
}

Response:

{
  "compressed_text": "Compressed text...",
  "original_tokens": 8432,
  "compressed_tokens": 324,
  "compression_ratio": 0.0384,
  "speedup": 26.0
}