Motivation

Problem

Terminal users often need fast model assistance without leaving the shell and without losing control over what actually runs on their machine.

Two common pain points:

• Generic chat UX does not map cleanly to shell workflows.
• Command generation can be unsafe unless execution is explicit and inspectable.

Design Goals

1. Local-first by default

ask is designed to work with local model backends through llm and tools like Ollama.

2. Shell-native ergonomics

The tool is installed as a normal CLI command and integrates with shell aliases and hotkeys.

3. Explicit safety boundaries

• -co generates commands only.
• -no executes generated commands only after a confirmation prompt.

4. Better answers when needed

The wrapper can inject retrieval context from web or local docs before model invocation.

5. Practical speed and feedback

Streaming output, spinner styles, and optional streaming TTS keep interaction responsive.

Why This Approach

This project favors practical terminal productivity over heavy framework complexity:

• minimal dependencies
• inspectable scripts
• deterministic shell setup
• explicit flags for powerful behavior