Run a model

conifer run --model <local-model-name> "explain a KV cache in two sentences"
git diff | conifer run --model <local-model-name> "write the commit message"

Name a model you pulled with models pull and it runs on this machine. The task comes from the arguments or from stdin. A bare conifer run in a terminal opens a chat session and keeps the model loaded until you leave.

Context windows#

A prompt longer than the model can hold is refused up front, with the model, its window and the prompt size named, rather than failing partway through. --ctx sets the window if you want a smaller one to save memory.

The engine#

Conifer runs local models on its own engine, built for Apple Silicon and for Windows. It fits the model to the memory you have, reuses what it has already computed across turns, and needs no configuration. Per-model speeds are on the models page.

Keeping a model loaded#

A one-shot run loads the model and lets it go when the run ends. A chat session keeps it. To keep models loaded for several tools at once, use serve.