gptcgt logogptcgt
Sign In

Getting Started

Get up and running with gptcgt in under 2 minutes.

System Requirements

  • Python 3.11+ — Required for the core application
  • macOS, Linux, or WSL — Full terminal support (Windows native coming soon)
  • At least one AI provider API key — Or sign up for Managed Credits

Installation

We recommend pipx for a clean, isolated install:

# Recommended
pipx install gptcgt

# Alternative: standard pip
pip install gptcgt

Launch

Navigate to any project directory and start gptcgt:

cd ~/my-project
gptcgt

gptcgt will automatically detect your project's language, test framework, and linter. It creates a .gptcgt/ directory inside your project for configuration, phase tracking, and agent memory.

First-Run Onboarding

The first time you launch gptcgt, the onboarding wizard walks you through:

  1. Authentication — Choose between BYOK (Bring Your Own Keys) or Managed Credits
  2. API Key Entry — If BYOK, enter keys for the providers you want (OpenAI, Anthropic, Google, etc.)
  3. Quality Tier — Set your default tier (Standard is recommended to start)
  4. Theme Selection — Pick a color theme for the terminal UI
  5. Terms of Service — Accept the ToS to proceed

You can re-run the wizard at any time with /setup.

The Interface

gptcgt presents a three-panel terminal layout:

  • Left Panel — File Tree: Your project files with real-time change indicators
  • Center Panel — Code Viewer: View and inspect files with syntax highlighting
  • Right Panel — Chat: Talk to the AI, view diffs, and approve changes

Toggle panels with Ctrl+B (files), Ctrl+J (chat). Press Tab to cycle focus between panels.

Your First Task

Focus the chat input (Tab or Ctrl+J) and type a natural language request:

Add input validation to the signup form in auth.py

gptcgt will:

  1. Analyze intent — Determine this is an “edit” task with complexity ~4/10
  2. Route to the best model — Based on complexity, your tier, and ELO ratings
  3. Gather context — Find relevant files using AST maps and symbol references
  4. Generate changes — Stream the response with a unified diff
  5. Present for approval — You review the diff and accept or reject it
  6. Security scan — Every change is scanned for vulnerabilities before application

What Happens Behind the Scenes

When you submit a task, it flows through a DAG (Directed Acyclic Graph) pipeline:

InitAnalyze → GatherContext → RouteTask → PrepareBlackboard
    ↓
[ArchitectPlan] or [StandardExecution] or [ParallelExecution]
    ↓
Arbiter Verification → Security Scan → Present to User

Each node in the pipeline is traceable — you can see timing data and transitions in the log panel.

Next Steps