A local AI skill for running and routing Microsoft's high-efficiency Phi model family across minimal hardware fleets.
Install & Download
1. ClawHub CLI
The fastest way to install a skill directly from the registry.
npx clawhub@latest install phi-phi4
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 phi-phi4 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 Phi 4 Microsoft Small Models?
The Phi 4 skill brings Microsoft's state-of-the-art small language models (SLMs) to your local environment through the Openclaw Skills ecosystem. These models are engineered to provide reasoning, math, and coding capabilities that rival much larger LLMs while maintaining a tiny memory footprint. This makes them ideal for developers running AI locally on consumer-grade hardware like MacBook Airs or Mac Minis without relying on expensive cloud APIs.
By leveraging the ollama-herd backbone, this skill allows you to distribute Phi models across your device fleet. Whether you are using the 14B Phi-4 for complex reasoning or the ultra-fast 3.8B Phi-4-mini for basic summarization, you get low-latency, private, and cost-effective AI performance across macOS, Linux, and Windows.
Phi 4 Microsoft Small Models Use Cases
- Running complex reasoning and math tasks on hardware with as little as 8GB RAM.
- Deploying a private, OpenAI-compatible API for local development and testing.
- Offloading AI inference to a fleet of low-power devices to increase total throughput.
- Implementing fast, local text summarization and unit test generation without data leaving your network.
How Phi 4 Microsoft Small Models Works
- The user initiates the skill, which acts as a bridge to the ollama-herd orchestration layer.
- A central router (herd) is started on port 11435 to manage incoming requests.
- Individual devices join as nodes (herd-node) and register their available hardware resources.
- When a request for a Phi model is made, the router identifies the best-suited device in the fleet based on available RAM and latency.
- The selected Phi model processes the prompt and returns the result through an OpenAI-compatible endpoint.
Phi 4 Microsoft Small Models Setup
To get started with these Openclaw Skills, ensure you have python3 and pip installed. Install the fleet manager and start the routing services using the following commands:
pip install ollama-herd
# Start the central router
herd
# Run this on each device in your local network
herd-node
Note: All model downloads require explicit user confirmation to ensure bandwidth and storage are managed intentionally.
Phi 4 Microsoft Small Models Data Schema & Taxonomy
The skill manages its operational data and logs within a dedicated fleet manager directory. Below is the primary data structure:
| Path | Purpose |
|---|---|
~/.fleet-manager/latency.db |
Stores network latency metrics between fleet nodes. |
~/.fleet-manager/logs/herd.jsonl |
Structured logs of model requests, performance, and system events. |
http://localhost:11435/v1 |
OpenAI-compatible API base URL for model interactions. |
All model files are managed via the underlying Ollama service, ensuring isolation from your primary system files.
Phi 4 Microsoft Small Models Advanced Features
- Multi-agent support for routing specific tasks to different versions of Phi simultaneously.
- Real-time fleet health monitoring via a web-based dashboard at port 11435.
- Automated model recommendations based on the specific RAM and GPU specs of connected hardware.
- Support for image generation (z-image-turbo) and speech-to-text (qwen3-asr) within the same fleet infrastructure.
SKILL.md
Loading
Related Openclaw Skills
Air Duct Cleaning Video
An automated video assistant designed to transform HVAC inspection footage into compelling before-and-after marketing clips for air duct and dryer vent cleaning services.
Private AI (Ollama Herd)
Run powerful AI models entirely on your own hardware without cloud dependencies, telemetry, or third-party access.
Local Coding Assistant
A distributed local inference engine that routes code generation, review, and refactoring tasks across your personal hardware fleet without cloud dependencies.
Baoyu Writing Style Skill
A structured methodology for technical writing that uses core metaphors and iterative refinement to create high-impact, actionable content.
Mistral Codestral Fleet Manager
Run and route the full suite of Mistral AI models across your local hardware fleet using a unified, OpenAI-compatible API.
Along Plan
A read-only exploration and planning protocol that enforces safe code analysis before any modifications are made.
Featured*
