frequently asked
Straight answers about what Mullama is, how it relates to Ollama and llama.cpp, and what runs where.
+ What is Mullama?
Mullama is a local LLM runtime written in Rust on top of llama.cpp. It runs GGUF models from Hugging Face, ships native in-process bindings for Rust, Python, Node.js, Go, PHP, and C/C++, is drop-in CLI-compatible with Ollama, and exposes OpenAI- and Anthropic-compatible HTTP APIs. MIT licensed.
+ Is Mullama a drop-in replacement for Ollama?
For the CLI and HTTP surface, yes. Mullama uses the same CLI verbs, Modelfile format, model registry, and port 11434, so existing Ollama client code keeps working. It adds native language bindings and an Anthropic-compatible API that Ollama does not have.
+ Which languages have native bindings?
Rust, Python, Node.js, Go, PHP, and C/C++. These are real in-process bindings to one Rust core, not HTTP wrappers — there is no daemon or subprocess between your app and the model.
+ Do the bindings require a running daemon?
No. Library mode links the same engine directly into your process, so you can generate without any daemon. The daemon exists for CLI use and for serving the HTTP APIs.
+ Which models does Mullama run?
Any GGUF file from Hugging Face, with pre-configured aliases for Llama, Qwen 2.5, DeepSeek R1, Mistral, Phi, Gemma, LLaVA vision models, and BGE/Nomic embeddings. Use a short alias, hf:owner/repo, or a path to a .gguf file.
+ What GPU backends are supported?
Seven: CUDA, Metal, ROCm, Vulkan, OpenCL, SYCL, and RPC for distributed inference across machines. Apple Silicon uses Metal automatically, and CPU works out of the box.
+ Which HTTP APIs does it speak?
OpenAI-compatible at /v1 and Anthropic-compatible at /v1/messages, both on port 11434. The openai and anthropic SDKs, LangChain, and LlamaIndex all work pointed at localhost.
+ Does Mullama send my data anywhere?
No. It runs on your machine or your server. There is no hosted service, no account, and no telemetry leaving the box unless you explicitly point the OpenAI/Anthropic endpoints at a remote backend.
+ Can I embed Mullama in a desktop or mobile app?
Yes. Link the static C ABI library into an Electron, Tauri, or native iOS/Android build and ship inference inside the app binary — no daemon, no network dependency.
+ Is Mullama a model trainer?
No. Mullama runs GGUF models for inference. Training and fine-tuning live elsewhere in the ecosystem, and it is not a SafeTensors runtime.
+ What license is Mullama under?
MIT. You can use it in commercial products without restriction. The source lives at github.com/cognisoc/mullama.
Still have a question about the runtime, a binding, or a model?