Skip to content

tui.md

Latest commit

 

History

History
87 lines (66 loc) · 3.59 KB

tui.md

File metadata and controls

87 lines (66 loc) · 3.59 KB
Raw

TUI Guide

Model CLI uses Huh for prompts, Bubble Tea for the interactive context panel, and Lip Gloss for terminal styling. The interactive Quick Test Drive is in the repository README; this page describes the terminal experience itself.

Wizard

Start the guided flow with:

model-cli wizard

The wizard follows these seven phases:

  1. Develop & Package
  2. Local Hardening & Compliance
  3. Supply Chain Check
  4. Manifest-Level Validation
  5. GitOps Admission & Policy Enforcement
  6. Infrastructure & Resource Orchestration
  7. Runtime Execution & Optimization

At context-panel pauses, the line above the panel states exactly what Enter will do next. For example, after model details, Enter starts local packaging; after packaging, Enter starts SBOM generation and MOF classification.

The wizard executes local packaging, hardening, local compliance, optional OCI publication, and manifest retrieval. Signing and verification are demonstrated in the wizard; use standalone sign and verify commands to execute them. When no Kubernetes cluster is connected, GitOps, infrastructure, and runtime phases are guided simulations.

Context Panel

The panel appears at phase transitions and has six tabs:

Tab Contents
Progress Current phase, progress bar, and completed, current, pending, or skipped phases
Config Selected registry, GitOps, signer, and runtime values
Model Model name, source path, artifact reference, and completed operation status
Logs Recent workflow events
Env Model CLI version, terminal detection, and Go version information
Help Keyboard controls

Use Tab and Shift+Tab to change tabs, or 1 through 6 to select a tab. Press Enter to continue the displayed workflow action. Press Esc to cancel the wizard.

The progress bar reflects the seven-phase journey. A check mark means the phase completed, an arrow marks the current phase, and a dot is pending. The panel also supports a dash for phases explicitly marked as skipped.

Prompt Behavior

Values supplied through command-line flags take precedence. Otherwise, the CLI uses a saved value from ~/.model-cli.yaml when one exists, or prompts for a choice. An explicit empty flag value is preserved rather than prompting again.

Prompts are disabled in non-interactive shells, with --non-interactive, or when MODEL_CLI_NO_INTERACTIVE=1. Required missing values then produce an error naming the flag to provide. The wizard command requires an interactive terminal.

Tool menus show supported choices and mark the current recommendation. Some choices are mutually exclusive providers, such as ORAS versus ModelPack; the serving-topology demonstration shows direct vLLM, KServe, and KServe-managed vLLM. When KServe-managed vLLM is selected, the wizard prompts for a Hugging Face model ID (default: facebook/opt-125m) to deploy. Note: vLLM serving requires GPU nodes and significant memory; the bundled CPU-only minikube cannot complete a vLLM serve.

Styling And Accessibility

The wizard uses white bold headers, blue step labels, green success messages, light-blue information, and orange warnings. Colors supplement text labels and symbols rather than carrying meaning by themselves.

Wizard styles are defined in cmd/wizard.go. Context-panel rendering and controls are implemented in internal/tui/context_tui.go.

Use a terminal with at least 100 columns and 24-bit color support for the most readable layout. VS Code, iTerm2, and Alacritty meet this requirement.