A powerful CLI skill for running, serving, and benchmarking local AI models using containerized runtimes like Docker and Podman.
Install & Download
1. ClawHub CLI
The fastest way to install a skill directly from the registry.
npx clawhub@latest install ramalama-cli
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 ramalama-cli 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 Ramalama CLI AI Agent Skill?
The Ramalama CLI skill enables developers to manage the entire lifecycle of local AI models with ease. By leveraging Openclaw Skills, users can execute containerized inference, ensuring that sensitive data never leaves their local environment. This tool acts as a bridge to various model sources, including Hugging Face, OCI registries, and direct URLs, providing a unified interface for model discovery and execution.
Whether you need to solve simple tasks with a lightweight local agent or deploy specialist models with unique capabilities, this skill provides the necessary infrastructure. It supports multiple runtimes such as llama.cpp, vLLM, and MLX, making it a versatile choice for developers seeking high-performance, private AI solutions within the Openclaw Skills ecosystem.
Ramalama CLI AI Agent Skill Use Cases
- Running local LLMs to process sensitive or private data without cloud exposure.
- Creating Retrieval-Augmented Generation (RAG) packages from local documentation for context-aware queries.
- Serving OpenAI-compatible API endpoints to power local development tools and applications.
- Benchmarking model performance and perplexity to select the best agent for specific hardware.
- Managing model lifecycles including pulling from Hugging Face, converting formats, and pushing to registries.
How Ramalama CLI AI Agent Skill Works
- The skill begins with a preflight check to ensure the ramalama binary and a container engine like Podman or Docker are active.
- Users specify a model source using prefixes like hf:// for Hugging Face or oci:// for container registries.
- The skill pulls or converts the model and initializes the specified runtime environment (e.g., llama.cpp).
- For inference tasks, the command is executed as either a one-shot prompt, an interactive session, or a background service.
- If RAG is required, the skill packages local directories into a knowledge bundle that the agent references during execution.
- Outputs can be requested in JSON format to facilitate further automation within Openclaw Skills workflows.
Ramalama CLI AI Agent Skill Setup
To get started with this skill, you must have ramalama and a container engine installed. This Openclaw Skills component supports both Homebrew and uv for installation.
# Install via Homebrew
brew install ramalama
# Or install via uv
uv tool install ramalama
# Verify the installation and container engine
ramalama version
podman info || docker info
Ramalama CLI AI Agent Skill Data Schema & Taxonomy
The Ramalama CLI skill organizes its operations through a clear taxonomy of models, runtimes, and local storage. Data is managed as follows:
| Component | Details |
|---|---|
| Model Store | Local directory for cached models, configurable via the --store flag. |
| RAG Packages | Knowledge bundles generated from local file paths or web URLs. |
| Benchmarks | Historical performance data stored for model comparison. |
| Runtimes | Specific engines used for inference (llama.cpp, vLLM, or MLX). |
| OCI Registries | Support for pulling and pushing models as container images. |
Ramalama CLI AI Agent Skill Advanced Features
- Support for detached services using the
servecommand to provide persistent OpenAI-compatible APIs. - Advanced RAG capabilities to bundle entire documentation directories for local agent querying.
- Model evaluation tools including
benchandperplexityfor quantitative quality assessment. - Granular hardware control with flags for engine selection (podman vs docker) and context size adjustment.
- Seamless integration with Hugging Face and other external model registries for rapid testing of new architectures within Openclaw Skills.
SKILL.md
Loading
Related Openclaw Skills
Salai MCP Israeli Grocery Assistant
An Israeli grocery shopping and price-comparison assistant utilizing the Salai MCP endpoint for seamless product discovery and cart management.
daily.dev API for AI Agents
Bridge the gap between LLM knowledge cutoffs and real-time tech trends using community-validated developer content.
DDGS Web Search Skill
A privacy-focused metasearch skill for AI agents to fetch real-time text, images, news, and videos without requiring an API key.
What If Factory
A creative brainstorming engine that transforms simple hypothetical questions into detailed, multi-format parallel universe narratives.
gogifocus
A versatile command-line interface for managing Google Workspace services like Gmail, Sheets, and Drive via OAuth.
Stripe Best Practices
A technical knowledge skill focused on architecting secure, modern, and compliant Stripe payment integrations using the latest API standards.
Featured*
