midnightInstall once.
The runner and its runtime resources live in ~/.midnight. Your source checkout stays separate.
LOCAL MODEL RUNNER BETA
Your models.
Your machine.
Run AI on your own hardware.
A native model runner, one familiar API.
INSTALL MIDNIGHTmacOS / Linux
curl -fsSL https://midnightrun.sh/install.sh | bashLatest release · Apple silicon, macOS 26+ · Asks before adding to PATH.
Apple silicon MetalLinux source build
A smaller path from model to API 01 / FROM TERMINAL TO INFERENCE
Keep it local.
Keep it local.
Keep it simple.
Bring your model.
Add a compatible MLX checkpoint to ~/.midnight/models, then start the runner.
$ midnight --model MODEL_NAMEUse a familiar API.
Connect your tools to Midnight’s local, OpenAI-compatible chat endpoint.
localhost:8080/v1/chat/completionsBUILT TO RUN
Read the documentation The runner is the foundation.
Studio and other tools stay separate. Midnight does the inference.