Capability · Framework — agents
GPT-Engineer
GPT-Engineer is the spiritual ancestor of today's autonomous coding agents. You write a short prompt in `prompt.md`, run the CLI, and the agent scaffolds a project — asking clarifying questions, writing multiple files, and iterating on feedback. It popularised the pattern later refined by Cursor, Aider, Devin, and Claude Code.
Framework facts
- Category
- agents
- Language
- Python
- License
- MIT
- Repository
- https://github.com/gpt-engineer-org/gpt-engineer
Install
python -m pip install gpt-engineer
export OPENAI_API_KEY=sk-...Quickstart
mkdir my-project && cd my-project
echo 'A CLI snake game in Python with unit tests' > prompt
gpte .
# for iterative improvement on an existing project:
gpte . -iAlternatives
- Aider — chat-based pair programmer
- OpenHands (formerly OpenDevin) — autonomous coding agent
- SWE-agent — research-oriented
- Claude Code / Cursor — commercial IDE-native agents
Frequently asked questions
Does GPT-Engineer run tests?
The default flow writes tests but does not execute them. The `--execute` flag runs a shell command (usually `pytest`) between iterations for feedback-driven refinement.
How does it compare to modern coding agents like Claude Code?
GPT-Engineer is smaller, simpler, and batch-oriented — great for learning. Modern agents like Claude Code add interactive editing, test execution, and multi-turn context management.
Sources
- GPT-Engineer GitHub — accessed 2026-04-20
- GPT-Engineer — docs — accessed 2026-04-20