A distributed local inference engine for running Google's Gemma 3 models across a fleet of devices with automatic load balancing.
Install & Download
1. ClawHub CLI
The fastest way to install a skill directly from the registry.
npx clawhub@latest install gemma-gemma3
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 gemma-gemma3 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 Gemma 3 Local Fleet Router?
Gemma 3 represents the pinnacle of Google's open-source model family, offering a massive 128K context window and robust performance across 140+ languages. This Openclaw Skills integration utilizes the Ollama Herd router to distribute the computational load across your local network, allowing you to run powerful models like Gemma 3 27B by leveraging the combined RAM of your available hardware.
By transforming your local machines into a unified inference cluster, you gain access to state-of-the-art coding and reasoning capabilities without per-token fees. The system is designed for privacy-conscious developers who need high-performance LLMs that stay entirely within their own network infrastructure.
Gemma 3 Local Fleet Router Use Cases
- Processing massive codebases or long-form technical documentation using the 128K context window.
- Deploying multilingual applications that require support for over 140 languages without cloud latency.
- Offloading heavy LLM inference from a primary development machine to a secondary Mac Studio or Linux server.
- Building private AI agents that require high-quality reasoning via Gemma 3 27B or CodeGemma variants.
How Gemma 3 Local Fleet Router Works
- The user starts the central fleet router using the Openclaw Skills infrastructure on a primary local machine.
- Individual nodes (other computers on the network) connect to the router, reporting their available hardware resources.
- When a request for a specific Gemma 3 model is received, the fleet router identifies the best available machine based on RAM and current load.
- If the model is not present, it triggers a user-confirmed pull to download the necessary weights via Ollama.
- The router handles the API request, manages the inference lifecycle, and returns the response through a unified OpenAI-compatible endpoint.
Gemma 3 Local Fleet Router Setup
To get started with this Openclaw Skills integration, follow these steps:
# Install the fleet management tool
pip install ollama-herd
# Start the central router on your main machine
herd
# Start a node on any additional devices in your network
herd-node
Gemma 3 Local Fleet Router Data Schema & Taxonomy
The skill manages its local state and logs within a dedicated fleet manager directory.
| File/Path | Description |
|---|---|
~/.fleet-manager/latency.db |
SQLite database tracking node response times and performance metrics. |
~/.fleet-manager/logs/herd.jsonl |
Structured JSON lines logs for tracking requests and fleet events. |
Ollama Models |
Model weights are stored in standard Ollama directories, managed via explicit user confirmation. |
Gemma 3 Local Fleet Router Advanced Features
- Multi-node load balancing that automatically routes requests to the machine with the most available VRAM.
- OpenAI SDK drop-in replacement support, allowing any existing AI tool to use local Gemma 3 models by changing the base URL.
- Native support for specialized variants like CodeGemma for high-fidelity programming tasks.
- Real-time fleet monitoring through a built-in web dashboard located at port 11435.
SKILL.md
Loading
Related Openclaw Skills
MLX Apple Silicon Local AI
A comprehensive suite for running high-performance, MLX-native AI models across a fleet of Apple Silicon devices.
homelab-ai
Transform your collection of spare machines into a high-performance local AI inference cluster without the complexity of Docker or Kubernetes.
Felo Mindmap
Generate professional AI mindmaps and structured thinking maps directly from text prompts using the Felo API.
A-Share Stock 10D Analysis
A comprehensive stock diagnostic tool that evaluates Chinese A-shares across ten critical dimensions using a 1000-point scoring system.
A-share Financial Analysis
A professional-grade tool for comprehensive China A-share financial diagnostics and automated HTML report generation via remote AI agents.
Shipping Booking Extractor
An AI-powered logistics tool that transforms unstructured shipping booking notes and instructions into standardized JSON data.
Featured*
