CLI reference

Ollaya is a single binary. If you have used Ollama, the commands will feel familiar.

Command	What it does
`ollaya serve`	Start the server on `127.0.0.1:11435`
`ollaya run MODEL [STATE]`	Answer questions about a state; pulls the model first if needed
`ollaya pull MODEL`	Download a model from the registry
`ollaya list`	List models on this machine
`ollaya ps`	List models loaded in memory
`ollaya show MODEL`	Show a model's details, questions and license
`ollaya stop MODEL`	Unload a running model
`ollaya rm MODEL`	Delete a model
`ollaya cp SOURCE DESTINATION`	Copy a model under a new name
`ollaya create NAME -f Modelfile`	Create a model from a Modelfile
`ollaya push MODEL`	Upload a model to a registry

Model names

Models are referenced as name:tag. When the tag is left out, latest is used:

ollaya run laya                 # same as laya:latest
ollaya run laya:multilingual
ollaya pull laya:en-fp32

Tags without a precision suffix resolve to fp16 on a GPU and fp32 on CPU. Add -fp16 or -fp32 to choose explicitly.

Starts the server that the CLI and your applications talk to. It listens on 127.0.0.1:11435 and serves both the native API (/api/*) and the TypeSafe-compatible API (/v1/*). See the API reference.

ollaya serve

ollaya run

Runs a model against a state and prints the answers. The model is pulled automatically if it is not on this machine yet.

ollaya run laya --preset triage "I was charged twice for my subscription this month."

--preset NAME uses a built-in question set such as triage. Models created from a Modelfile with QUESTIONS already know what to ask, so the state is all you pass:

ollaya run triage "My invoice shows the wrong company name."

After a request, the model stays loaded for keep_alive (five minutes by default), so the next request skips the load.

ollaya pull

Downloads a model and verifies every layer.

ollaya pull laya

ollaya list and ollaya ps

list shows models on disk with their size and when they were modified. ps shows models loaded in memory and how long they will stay loaded.

ollaya show

Prints a model's details: backbone, context length, precision, baked-in questions, calibration and license.

ollaya show laya:en

ollaya stop

Unloads a running model right away instead of waiting for keep_alive to expire.

ollaya rm and ollaya cp

ollaya cp laya:en my-guardrail
ollaya rm my-guardrail

ollaya create

Builds a derived model from a Modelfile, for example to bake a question schema or a refit calibration into a model.

ollaya create triage -f Modelfile

ollaya push

Uploads a model you created to a registry, so others can pull it.