What it does
GPT Engineer is a Python command-line program for experimenting with code generation. You write instructions in a plain file called prompt inside a project folder and run the gpte command on that folder, and the AI writes and runs the code. Running it with the -i flag on a folder of existing code asks it to improve that code instead. Out of the box it uses OpenAI or Anthropic models through an API key set in the environment or a .env file, and it can be configured for Azure OpenAI or for local and open-source models with extra steps. Other options include image inputs for vision-capable models, custom pre-prompts that change the agent's identity, and a bench command for testing your own agent implementations against public datasets such as APPS and MBPP. The repository was archived by its owner on April 22, 2026 and is now read-only. Its README describes gptengineer.app as a separate commercial product that generates web apps, and recommends aider as a well maintained, hackable CLI.
Capabilities
- Builds a project from instructions in a plain prompt file
- Improve mode for existing code with the -i flag
- OpenAI and Anthropic models by default; local or open-source models with extra setup
- Image inputs for vision-capable models
- Custom pre-prompts to change the agent's behavior across projects
- bench command for testing your own agents on APPS and MBPP
When to use it
- Trying prompt-to-project code generation from a terminal
- Asking an AI to draft improvements to a folder of existing code
- Benchmarking your own coding agent against public datasets
- Reading how an early open-source code-generation CLI worked, keeping in mind the repository is archived
Alternatives to explore
- Aider
