A serious model. On the Mac you own.

Rizz runs large open weight models on Apple Silicon, using the Metal and MLX backends built for it. Everything stays on your machine, and everything that talks to a cloud model can talk to this instead.

What you can do with it

• Serve: run a model on your Mac and point anything at it, from coding agents to your own scripts.
• Write: rewrite what you have written and write what you have not, in every app on your Mac.
• Chat: talk to the model in the terminal without leaving the window you are in.
• Code: a coding session that reads and edits your workspace against the model on your machine.
• Measure: find out what your Mac does with a model, in tokens per second, before you decide anything.
• Manage: convert checkpoints, use models your other tools downloaded, and see what a server is doing.

Measured

On an Apple M4 Max with 36 GB of unified memory, Qwen3 4B in native MLX Q3 affine decodes at 126 to 129 tokens per second on the Rizz Metal runtime, against about 105 on Ollama for the same checkpoint.

Read about Rizz Write