@thomasrumas/llm-client

CLI client to list, launch, and stop LLM inference servers managed by @thomasrumas/llm-manager over your local network.

Requirements

• Node.js 22+
• A machine on your LAN running @thomasrumas/llm-manager with the API server available — either:
  • Daemon (recommended): llm-manager service install && llm-manager service start on the remote machine
  • Embedded: API server enabled in the TUI Settings (runs while the TUI is open)

Install

npm install -g @thomasrumas/llm-client

Quick start

# 1. Point the client at your manager machine
llm-client config set remote-url http://192.168.1.5:3333

# 2. See what's available
llm-client list

# 3. Launch a model
llm-client start Qwen3-8B

# 4. Check it's running
llm-client status

# 5. Stop it
llm-client stop

Features

• Zero dependencies — uses Node's built-in fetch, no runtime packages
• Model name without .gguf — type Qwen3-8B instead of Qwen3-8B-Q4_K_M.gguf; the extension is added automatically
• Default model — configure a default so llm-client start always knows what to launch
• Named configurations — target any saved config with --config <name>
• 10 s request timeout with a clear "cannot reach server" message

Commands

config

llm-client config show
llm-client config set <key> <value>

| Key | Description | | ---------------- | -------------------------------------------------------------------- | | remote-url | Base URL of the manager API (e.g. http://192.168.1.5:3333) | | default-model | Model name to use when none is specified on start | | default-config | Config name to use when --config is omitted (default: "default") |

Client config is stored at ~/.local-llm-client/config.json.

list

llm-client list

Lists every model that has at least one saved configuration on the remote manager, along with their config names. Models without a configuration cannot be launched remotely.

start

llm-client start [model-name] [--config <name>]

Launches a model on the remote manager. If model-name is omitted, uses default-model from config. The .gguf extension is optional — both Qwen3-8B and Qwen3-8B-Q4_K_M.gguf are accepted.

llm-client start                          # uses default-model + default config
llm-client start Qwen3-8B                 # specific model, default config
llm-client start Qwen3-8B --config fast   # specific model + named config

status

llm-client status

Shows the currently running model (if any): name, config, port, PID, and uptime. Also prints the OpenAI-compatible endpoint URL.

stop

llm-client stop

Stops the model currently running on the remote manager.

help

llm-client help

How the manager API must be enabled

Option A — Daemon (recommended for always-on access)

On the remote machine:

llm-manager service install   # registers with launchd / systemd, starts at login
llm-manager service start     # starts now

The daemon runs the API server headlessly with no terminal window required. The TUI can still be used alongside it.

Option B — Embedded API inside the TUI

On the machine running the manager TUI:

1. Open Settings (4 from the dashboard)
2. Navigate to API Server and toggle it enabled with ← →
3. Set the desired API Port (default 3333)
4. Press Ctrl+S to save — the API server starts immediately

The API listens on 0.0.0.0 so it is reachable from any device on the same network.

License

ISC