CLI Command Reference

NERVI provides a robust suite of commands to launch the server, manage model files, configure nicknames, and interact with the NPU hardware.

🚀 Server Lifecycle Commands

serve

Runs the npuserver backend in the foreground. This is useful for debugging server logs in real time.

  • Syntax: nervi serve

kill

Stops any managed background server process running on the host.

  • Syntax: nervi kill
  • Mechanism: Reads the Process ID from %AppData%\Roaming\nervi\server.pid and terminates the process safely.

💾 Model Lifecycle Commands

load

Loads a pre-compiled model into NPU memory.

  • Syntax: nervi load <model-name-or-port>
  • Example: nervi load qwen3-4b

unload

Unloads the active model from NPU memory to free up hardware resources.

  • Syntax: nervi unload

delete

Removes compiled cache directories and weight files from the disk.

  • Syntax: nervi delete <model-id> [flags]
  • Flags:
    • -c, --compiled-only: Deletes only compiled NPU blobs but keeps downloaded weights.
  • Example: nervi delete durgasai299792458/Qwen3-4B-OpenVINO-INT4-npu-i

🩺 System and Diagnostic Commands

health

Fetches real-time server health and NPU telemetry.

  • Syntax: nervi health

info

Prints active configurations, connection URLs, ports, and configuration file paths.

  • Syntax: nervi info

update

Upgrades the python package dependencies in the managed environment.

  • Syntax: nervi update

🧩 Extension Commands

app list

Displays all installed extensions showing version numbers and current status.

nervi app list

app install <extension-id>

Downloads and installs an extension from the marketplace.

nervi app install <extension-id>

app create <extension-id>

Scaffolds a new extension project.

nervi app create <extension-id> [--mode=tool|menu] [--type=stateless|stateful]