documentation / introduction

How Ashplug works

Ashplug is an AI-powered shell assistant. It turns natural-language intent into a typed execution plan, then keeps validation, safety, execution, and history inside the Rust application.

Core boundaryThe AI proposes data. It never executes a command, changes policy, or approves its own operation.

Request lifecycle

  1. 01CLI receives a natural-language task.
  2. 02A provider returns a structured plan.
  3. 03The planner validates the plan.
  4. 04The safety engine returns allow, approval, or deny.
  5. 05Approved commands execute without a shell.
  6. 06Output and decisions are recorded in history.

What it supports

The current MVP supports OpenAI-compatible, Ollama, and mock providers; Bash and Zsh integration; configurable safety policy; command timeouts and bounded output; SQLite history; and bounded recovery after failed steps.