A comprehensive management suite for orchestrating Ollama models across multiple machines with AI-powered hardware recommendations.
Install & Download
1. ClawHub CLI
The fastest way to install a skill directly from the registry.
npx clawhub@latest install ollama-manager
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 ollama-manager 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 Ollama Manager?
Ollama Manager provides a centralized control plane for users running Ollama across a distributed fleet of hardware. It solves the common problem of model fragmentation and disk bloat by providing clear visibility into which models are loaded in VRAM, which are sitting idle on disk, and how much storage is being consumed across every node in your network. By integrating with Openclaw Skills, it allows developers to automate the lifecycle of Large Language Models (LLMs) with surgical precision.
The tool goes beyond simple listing by analyzing hardware specifications—including RAM, CPU cores, and GPU memory—to offer AI-powered recommendations for the optimal model mix. Whether you are managing a single Mac Studio or a diverse cluster of Linux and Windows machines, this skill ensures your local AI infrastructure remains lean, performant, and perfectly balanced for your specific hardware constraints.
Ollama Manager Use Cases
- Identifying and removing unused, high-capacity models to reclaim hundreds of gigabytes of disk space.
- Monitoring real-time VRAM usage to identify 'hot' models ready for instant inference versus 'cold' models requiring load time.
- Benchmarking model performance and latency across different hardware configurations to find the fastest node for specific tasks.
- Orchestrating model distribution across a multi-node 'herd' to ensure high availability and balanced resource utilization.
How Ollama Manager Works
- The system utilizes the
ollama-herdrouter as a central hub, typically running athttp://localhost:11435. - Individual
herd-nodeinstances run on each machine in the fleet, reporting heartbeats, loaded models, and disk telemetry back to the router. - User requests for model metadata or management actions are processed by the router, which queries the distributed nodes.
- Performance data and request traces are logged into a local SQLite database (
latency.db) for long-term usage analysis. - AI-driven recommendation logic compares hardware telemetry against model requirements to suggest the best deployment strategy.
Ollama Manager Setup
Install the management toolkit and initialize the router and nodes across your fleet:
# Install the Ollama management toolkit
pip install ollama-herd
# Start the main Ollama router on your primary machine
herd
# Run the node agent on every machine you wish to manage
herd-node
Ollama Manager Data Schema & Taxonomy
Ollama Manager organizes operational data within the ~/.fleet-manager/ directory. Key data structures include:
| Component | Path / Source | Description |
|---|---|---|
| Latency DB | ~/.fleet-manager/latency.db |
SQLite database tracking model requests, completion tokens, and per-node latency. |
| Logs | ~/.fleet-manager/logs/herd.jsonl |
JSONL formatted logs for system events and heartbeats. |
| Model Metadata | /api/tags |
Aggregated JSON output showing model versions, sizes, and node locations. |
| Hardware Profile | /api/recommendations |
Maps hardware specs (RAM, GPU) to recommended model quantizations. |
Ollama Manager Advanced Features
- Auto-pull Logic: Automatically downloads missing models to the node with the highest available disk and memory overhead.
- Thrashing Detection: Automated health checks that alert users when models are frequently loading and unloading, indicating VRAM pressure.
- Fleet Performance Analytics: Advanced SQL-based reporting to identify the most cost-effective and performant models based on actual token generation speeds.
- Centralized Dashboard: A web-based visual interface for one-click model deployment and real-time fleet health monitoring.
SKILL.md
Loading
Related Openclaw Skills
Markdown to DOCX Converter
A high-fidelity conversion engine that transforms Markdown documents into fully formatted, professional Word DOCX files.
24h Product Marketing Skill
An automated content generator that converts product URLs from 24h.chichbong.com into optimized, retail-ready Vietnamese product listings.
Venn Universal MCP Connector
Venn acts as a powerful architectural bridge connecting your AI agent to an entire ecosystem of enterprise SaaS tools through a single MCP gateway.
URL Toolkit
A comprehensive developer utility for encoding, decoding, parsing, and building URLs and query strings via a streamlined CLI.
JLCEDA Plugin Builder
A professional development toolkit for creating, debugging, and localizing high-performance extension plugins for the EasyEDA Pro ecosystem.
Placetel
A comprehensive communication skill that enables AI agents to manage Placetel cloud telephony, users, and business workflows through the Membrane ecosystem.
Featured*
