Skip to content

Docs Getting Started

Getting Started

Requirements

Before using ASDT, you need:

  • Claude Code or OpenCode — installed and authenticated (these are the only assistants ASDT targets today)
  • Engram — the memory provider, required for cross-session persistence. It must be on your PATH before you install
  • A terminal (bash or zsh)

Building from source? Go 1.22+ is required. Running the one-line installer downloads a pre-built binary — no compiler needed.

Installation

curl -fsSL https://raw.githubusercontent.com/vitualizz/asdt/main/install.sh | bash

Downloads the latest pre-built binary (asdt-tui) for your platform and installs it to ~/.local/bin/. No Go or compiler needed.

If ~/.local/bin is not on your PATH, the installer prints the exact line to add to your shell profile.

Run the installer

asdt-tui

asdt-tui is an interactive terminal UI. From the main menu, choose Install / Update Skills and it walks you through a short guided flow:

  1. Language — pick the language ASDT speaks in: en (English), es-419 (Latin American Spanish), or es-ES (peninsular Spanish). Your choice drives both the installer UI and the installed specialists.
  2. Environment check — probes your OS, shell, Engram, and Codegraph (an optional code-search index). Engram is a hard gate: Continue stays disabled until Engram is found on your PATH. A missing Codegraph is only a soft warning and never blocks you.
  3. Assistants — choose which to target: Claude Code, OpenCode, or both. Each shows a present/missing badge.
  4. Memory provider — Engram is the provider ASDT installs against.
  5. Model preset — pick how much model horsepower each specialist step uses (see below).
  6. Persona — optionally pick a persona for the installed assistants, and whether they should use emojis. You can skip this entirely.
  7. Permissions — an opt-in gate for the Claude Code permission overlay. Declining is the safe default and writes nothing.
  8. Review → Install → Done — confirm your selections, watch the install run, and you’re set. A read-only Dashboard on the main menu shows per-assistant status any time.

Model presets

The model step offers five presets — each maps every specialist step to a model tier:

  • Chameleon

    Default

    Keeps the model your assistant already has defined (strips the model: field so each assistant uses its own default).

  • Sprinter

    Fastest and cheapest across the board.

  • Craftsman

    Recommended

    The recommended balance of speed and capability (the shipped defaults, verbatim).

  • Strategist

    More capability for analysis and decisions.

  • Mastermind

    Maximum capability where it matters most.

Customize per step

Set the model for each specialist step yourself.

Initialize your project

Open your AI assistant in your project folder and run:

/asdt-init

It detects your stack, asks a couple of questions, and writes .asdt/config.yaml with sensible defaults.

Your first pipeline

/asdt Add user authentication with email and password

ASDT analyzes the request and recommends a specialist sequence — for example: /asdt-pm → /asdt-architect → /asdt-developer. Confirm the plan, then run each command. Each specialist saves its output to the knowledge base so the next one picks up where the previous left off.

Running individual specialists

/asdt-pm Add dark mode to the settings page
/asdt-architect Design the caching strategy
/asdt-developer Implement the user profile component

Next steps