CLI options

Run ./anna without arguments to display the usage synopsis. The -m option is required and selects the model file.

Option	Purpose
-m <file>	Load a model file.
-t <count>	Set the number of CPU threads.
-p <file>	Load a prompt file; can be used more than once.
-n <count>	Limit generated tokens per reply.
-e <temperature>	Set sampling temperature.
-x <tokens>	Set context length.
-G <layers>	Offload model layers to the GPU.
-V <file>	Load a vision projector file.
-i <file>	Add an image as a secondary prompt.
-R <URL>	Use an ANNA server for remote offloading.

Internal commands

At a user-input prompt, ANNA also accepts commands such as quit(), save(), load(), load_file(), no_input(), and image().

This page is an initial reference and will be expanded with every supported option and examples verified against the current release.