gotg 0.1.0

AI agents collaborating like real engineering teams

GOTG

GOTG is an AI product and engineering department. We've taken real-world experience working with high-performing engineering teams and distilled it into a development tool built for AI-assisted software development. Instead of treating AI as a tool inside your editor, GOTG treats AI agents as team members with roles, autonomy, and communication responsibilities. This isn't multiple agents working on isolated tasks — it's AI agents and humans collaborating like real engineering teams, following a structure that ensures higher quality and safer products.

Prerequisites

• Python 3.11+
• Git (required — gotg init requires a git repository)
• One of:
  • Anthropic API key (recommended, default — Claude Sonnet 4.6)
  • OpenAI API key (GPT-4.1 mini, GPT-4.1, o3, o4-mini)
  • Gemini API key (Gemini 2.5 Flash, Gemini 2.5 Pro)
  • Ollama running locally (free, no API key needed)

Install

git clone https://github.com/biff-ai/gotg.git
cd gotg
python3 -m venv .venv
source .venv/bin/activate
pip install -e .

# Optional: install the TUI
pip install -e ".[tui]"

Quickstart

mkdir my-project && cd my-project
git init -b main && echo "# my-project" > README.md && git add . && git commit -m "init"
gotg init .
gotg model anthropic
echo "ANTHROPIC_API_KEY=sk-ant-..." > .env
gotg new "Build a CLI todo list app"
gotg run

Agents discuss your task through five phases: refinement, planning, pre-code-review, implementation, and code-review. The coach signals when each phase is ready to advance. You control transitions with gotg advance.

For the full walkthrough, see the User Guide.

How It Works

GOTG puts you in the role of Product Manager. AI agents discuss and design based on your task description, guided through structured phases by an AI coach. Tasks are organized into dependency layers. Each layer cycles through: implementation → code-review → merge → next-layer.

See the User Guide for phases, layers, and conversation mechanics.

Commands

Core

Command	Description
gotg init [path]	Initialize .team/ in a git repo
gotg new "description"	Create a new iteration
gotg run [--max-turns N] [--layer N]	Start the agent conversation
gotg continue [-m MSG] [--max-turns N]	Resume with optional human input
gotg show	Replay the conversation log
gotg advance	Advance to the next phase
gotg model [provider] [model]	View or change model config

Worktrees & Merge

Command	Description
gotg worktrees	List active worktrees with status
gotg commit-worktrees [-m MSG]	Commit all dirty worktrees
gotg review [branch] [--layer N] [--stat-only]	Review agent diffs against main
gotg merge <branch | all> [--layer N] [--abort]	Merge agent branches into main
gotg next-layer	Advance to the next task layer (after merging)
gotg rework	Send tasks back for rework after changes_requested review

File Approvals

Command	Description
gotg approvals	Show pending file write requests
gotg approve <id | all>	Approve a pending write
gotg deny <id> [-m reason]	Deny a pending write with reason

Exploration

Command	Description
gotg explore start "topic" [--coach] [--slug S]	Start a new exploration
gotg explore continue <slug> [-m MSG] [--max-turns N]	Resume an exploration
gotg explore list	List all exploration sessions
gotg explore show <slug>	Replay an exploration conversation
gotg explore summarize <slug>	Generate summary from conversation

See User Guide — Exploration for context injection, read-only file tools, and TUI usage.

Terminal UI

gotg ui

Launches a Textual-based terminal interface for managing iterations, running sessions, and reviewing code. Requires pip install gotg[tui].

See the User Guide — TUI Walkthrough for screen descriptions and key bindings.

Development Setup

git clone https://github.com/biff-ai/gotg.git
cd gotg
python3.11 -m venv .venv
source .venv/bin/activate
pip install -e .
pip install pytest

gotg --help
pytest -q

The editable install (pip install -e .) means the gotg command points directly at source files. Edit code and re-run without reinstalling.

Running tests

pytest -q                     # all tests
pytest tests/test_worktree.py # just worktree tests
pytest -k "merge"             # tests matching "merge"

Supported Models

GOTG defaults to Anthropic Claude Sonnet 4.6. You can configure different providers per-team or per-agent using gotg model or the TUI settings screen.

Provider	Default	Available Models
Anthropic	claude-sonnet-4-6	claude-opus-4-6, claude-sonnet-4-5, claude-haiku-4-5, claude-opus-4-5, claude-sonnet-4, claude-opus-4, claude-3-7-sonnet, claude-3-5-haiku
OpenAI	gpt-4.1-mini	gpt-4.1, gpt-4.1-nano, gpt-5-mini, gpt-4o, gpt-4o-mini, o4-mini, o3
Gemini	gemini-2.5-flash	gemini-2.5-pro, gemini-2.0-flash, gemini-2.0-flash-lite, gemini-1.5-pro, gemini-1.5-flash
Ollama	qwen2.5-coder:7b	qwen2.5-coder:14b/32b, llama3.2:8b, deepseek-coder-v2:16b, codellama:13b

Per-agent model overrides let you mix providers — e.g. one agent on Claude, another on GPT-4.1. Configure in the TUI settings screen or directly in .team/team.json.

Using Ollama (local, free)

curl -fsSL https://ollama.com/install.sh | sh
ollama pull qwen2.5-coder:7b
ollama serve

For AMD GPUs (RX 6000/7000 series):

HSA_OVERRIDE_GFX_VERSION=10.3.0 ollama serve

Configure: gotg model ollama. No API key needed.

Note: Local models produce noticeably lower quality conversations than Anthropic/OpenAI.

Project Background

See narrative.md for the full design conversation and architectural decisions behind GOTG. See docs/ for the user guide and technical documentation.