TokenRanger is a local context compression plugin for Openclaw Skills that slashes cloud LLM token costs by up to 80% using local SLMs.
Install & Download
1. ClawHub CLI
The fastest way to install a skill directly from the registry.
npx clawhub@latest install tokenranger
2. Manual Installation
Copy the skill folder to one of these locations
~/.openclaw/skills/ <project>/skills/ Priority: Workspace > Local > Bundled
3. Prompt Installation
Copy this prompt to OpenClaw to install it automatically.
Help me install tokenranger using Clawhub. If Clawhub is not installed, install it first (npm i -g clawhub).
Prefer to download?
Get the raw skill files in a ZIP archive.
What is TokenRanger?
TokenRanger is a powerful performance-oriented plugin designed for the Openclaw ecosystem. It acts as a middleware between the user and cloud LLMs, intercepting conversation history and compressing it through a local Ollama instance before transmission. This significantly reduces input token usage while maintaining semantic accuracy, making it one of the most cost-effective Openclaw Skills for heavy developers.
By leveraging Small Language Models (SLMs) like Mistral or Phi-3 locally, TokenRanger ensures that your cloud bills remain low without sacrificing the intelligence of high-end models. It features graceful degradation, meaning if your local service is unavailable, it simply falls back to standard passthrough mode to ensure uninterrupted service.
TokenRanger Use Cases
- High-volume developers looking to slash API costs for GPT-4 or Claude 3.5 Sonnet
- Users working with long conversation histories that exceed standard context windows
- Privacy-conscious environments requiring local pre-processing of session data
- Systems operating on limited bandwidth where smaller prompt sizes are beneficial
How TokenRanger Works
- The OpenClaw gateway receives a user message and triggers the TokenRanger hook.
- For the first turn, the full context is sent to ensure high fidelity.
- From the second turn onwards, conversation history is sent to a local FastAPI sidecar.
- The sidecar utilizes Ollama to run a compression chain via LangChain LCEL.
- A semantically dense summary is generated and prepended to the current prompt.
- The cloud LLM receives the compressed context, drastically reducing the billable token count.
TokenRanger Setup
To integrate this into your library of Openclaw Skills, follow these steps:
Install the plugin:
openclaw plugins install openclaw-plugin-tokenranger
Run the initial setup to configure the sidecar and pull local models:
openclaw tokenranger setup
Restart your gateway to apply changes:
openclaw gateway restart
Verify the installation:
openclaw tokenranger
TokenRanger Data Schema & Taxonomy
TokenRanger manages its environment and configuration through several key files and services:
| Component | Path / Detail |
|---|---|
| Sidecar Service | systemd (Linux) or launchd (macOS) |
| Python Venv | ~/.openclaw/extensions/tokenranger/venv |
| Config File | ~/.openclaw/config.json |
| Local Models | mistral:7b (GPU) or phi3.5:3b (CPU) |
| Sidecar Logs | ~/.openclaw/extensions/tokenranger/service/logs |
TokenRanger Advanced Features
- Multi-Strategy Inference: Automatically switches between GPU-based semantic summarization and CPU-based extractive bullet points.
- Dynamic Context Control: Configure minimum prompt lengths via minPromptLength to prevent compression on short messages.
- In-Chat Commands: Real-time mode switching and status checks using /tokenranger commands directly in the interface.
- Robust Fallback: Silent failover to uncompressed context if the local Ollama service becomes unreachable, ensuring reliability.
SKILL.md
Loading
Related Openclaw Skills
Video to Text
A local-first AI transcription tool that generates text transcripts and SRT subtitles from media files without requiring an external API.
Video Resize
A local and AI-powered video reformatting tool designed to convert any video into platform-specific aspect ratios.
Video Clip
A high-performance media utility for instant local video trimming and AI-driven highlight extraction.
Flight Search Plus
A real-time flight search and price comparison tool that integrates global airfare data into AI workflows.
CLAW99 SDK
The CLAW99 SDK connects your AI agents to a decentralized marketplace on Base to compete for crypto bounties in categories like DeFi, coding, and security.
OpenSETI Distributed Scanner
OpenSETI is a distributed computing skill that allows users to scan real radio telescope data for extraterrestrial signals and earn token rewards.
Featured*
