Setup & Installation
Or with the ClawHub CLI, for registry-managed skill folders outside a full OpenClaw workspace:
What This Skill Does
ramalama runs AI models locally inside containers using Docker or Podman. It supports multiple model sources including HuggingFace, OCI registries, and direct URLs. Models can be run one-shot, served as an OpenAI-compatible endpoint, or queried with RAG context from local files.
Containerized execution via Docker or Podman gives full network and device isolation, which cloud-hosted inference APIs cannot provide.
When to use it
- Run inference on sensitive documents without sending data to external APIs
- Serve a local OpenAI-compatible endpoint for development and testing
- Build a RAG knowledge bundle from internal docs and query it with a local model
- Benchmark model quality and performance before integrating into a pipeline
- Pull and test a specific HuggingFace model variant without a cloud account
Example Workflow
Here's how your AI assistant might use this skill in practice.
User asks: summarize our internal auth docs using a local model
- 1Run preflight: ramalama version and verify Docker or Podman is available
- 2Package local docs into a RAG bundle: ramalama rag ./docs auth-rag
- 3Run one-shot inference with RAG context: ramalama run --rag auth-rag granite3.3:2b "Summarize the auth requirements"
- 4Parse the text output and return the summary
A concise summary of the auth requirements drawn from the local documentation
Requirements
Accounts, API keys, or tools you or your AI assistant may need to set up while using this skill.