CLI

cli/ is the Bun command-line client for a vLLM Studio controller. It is useful for quick status checks, model lifecycle commands, and terminal dashboards without opening the frontend.

What It Does

• Reads status and GPU information from a controller.
• Lists recipes and configuration.
• Launches and evicts models.
• Uses controller APIs, so runtime support follows the connected controller: vLLM, SGLang, llama.cpp, MLX, and configured recipe command overrides.
• Provides an interactive terminal UI and headless commands.

What Is In Use

• Bun runtime.
• TypeScript source compiled with bun build.
• Direct HTTP calls to the controller API.

Architecture

flowchart LR
    User["Terminal user"] --> CLI["cli/src/main.ts"]
    CLI --> Headless["headless command mode"]
    CLI --> TUI["interactive TUI mode"]
    Headless --> API["controller HTTP API"]
    TUI --> API
    API --> Runtime["controller-managed runtimes"]

Prerequisites

• Bun 1.x.
• A reachable controller. The default URL is http://localhost:8080.

Commands

bun install
bun src/main.ts status
bun src/main.ts gpus
bun src/main.ts recipes
bun src/main.ts config
bun src/main.ts metrics
bun src/main.ts launch <recipe-id>
bun src/main.ts evict
bun src/main.ts help

Interactive mode:

bun src/main.ts

Configuration

• VLLM_STUDIO_URL: controller base URL. Defaults to http://localhost:8080.

The CLI does not discover local runtimes itself. It asks the connected controller for status, recipes, model lifecycle actions, and metrics.

Development

bun run typecheck
bun run lint
bun run build
bun run check

The build command emits the compiled vllm-studio CLI binary.

Where To Look

• src/main.ts: entrypoint.
• src/headless.ts: headless commands.
• src/api.ts: controller HTTP client.
• src/views/: interactive TUI views.