Llama Cpp Plugin: Managed Local GGUF Chat and Embeddings
This page covers the Llama Cpp plugin for OpenClaw, a locally managed llama.cpp server for GGUF chat and embeddings. It details installation, default models, and memory requirements for developers and users.
Read this when
- You are installing, configuring, or auditing the llama-cpp plugin
Llama Cpp plugin
A locally managed llama.cpp server that handles GGUF chat and embedding workloads.
Distribution
- Package:
@openclaw/llama-cpp-provider - Install route: npm; ClawHub
Surface
providers: llama-cpp; contracts: embeddingProviders
Default text model
When you go through interactive setup, OpenClaw pulls in a pinned, verified llama-server
and presents Gemma 4 E4B IT Q4_K_M as an approximately 5.0 GB download. That model option needs a minimum of 16 GiB of total RAM. Even on machines with less memory, any previously cached models are still recognized.
If you want a different model, point params.modelPath at any custom GGUF file. Those custom models bypass the RAM constraint tied to the bundled download. For systems that fall short of the requirement, a smaller model can be run through Ollama or LM Studio, or you can opt for a cloud provider.