AntonOsika

gpt-engineer: Generate and Improve Code from Natural Language

gpt-engineer is a Python CLI for turning natural-language specifications into code and iterating on existing projects with AI. It suits developers experimenting with code-generation agents who want a hackable local workflow.

View on GitHub Compare
Stars
55.1k
Forks
7.3k
Stars, last 30 days
-70

Our take

Health score

38/100

How it is scored
Activity
19
Community
90
Issues
21
Pull requests
20

-70 stars in the last 30 days

  • More contributors than 78% of the projects we track

Project history

Daily GitHub snapshots since Oct 7, 2026. History covers stars and forks; health signals are not tracked historically.

Selected snapshot

Stars

54,955

Forks

7,259

Change since Nov 16, 2026: 0 stars·-3 forks

Overview

gpt-engineer is a command-line platform for experimenting with AI code generation. Give it a natural-language specification, and it can write and execute code; it can also take instructions to improve an existing codebase. The project is aimed at developers exploring coding agents rather than people seeking a managed, non-technical app-building service.

It supports several model providers and can use image inputs with vision-capable models, making it useful for prototyping workflows where developers want to control prompts and agent behavior. The repository describes itself as an early code-generation experimentation platform and points users seeking a maintained, hackable CLI toward aider.

Key Features

  • Generate project code from instructions in a prompt file.
  • Ask the agent to modify an existing project using improvement instructions.
  • Customize agent identity and persistent guidance with pre-prompts.
  • Provide image inputs, such as UX or architecture diagrams, to vision-capable models.
  • Use OpenAI, Azure OpenAI, Anthropic, or configured open-source and local models.
  • Benchmark custom agent implementations against APPS and MBPP datasets through the bench command.
  • Run from a terminal, in Docker, or through GitHub Codespaces.

Use Cases

  • Developers can quickly turn a written product idea into a code prototype and inspect the generated result.
  • Coding-agent builders can test custom prompts or agent implementations against benchmark datasets.
  • Teams can experiment with agent workflows that incorporate diagrams or other visual project context.
  • Developers maintaining an existing project can provide targeted instructions and ask an AI to implement changes.

Getting Started

Install the stable release and set an OpenAI API key:

python -m pip install gpt-engineer
export OPENAI_API_KEY=[your api key]

Create a project folder with a prompt file, then run gpte <project_dir>. For more setup options and usage details, see the README.

Alternatives

  • pi: Pi is a TypeScript agent toolkit and CLI, while gpt-engineer is a Python CLI focused on generating and iterating on projects from specifications.
  • opencode: OpenCode provides a coding-agent workflow with plan and build modes plus terminal and desktop interfaces, rather than gpt-engineer’s spec-driven project generation.
  • aider: Aider focuses on making changes within existing repositories and integrating edits with version control, while gpt-engineer also generates projects from specifications.
  • Codebuff: Codebuff uses multiple agents to edit an existing codebase from terminal instructions, while gpt-engineer emphasizes generating projects from natural-language specifications.
ProjectLanguageLicenseStarsStatus
gpt-engineerPythonMIT55.1kNot checked yet
PiTypeScriptMIT117kActive
opencodeTypeScriptMIT212kActive
aiderPythonApache-2.049.4kQuiet
freebuffTypeScriptApache-2.013.1kActive

Considerations

  • The repository is archived, so users should account for limited future maintenance before adopting it for ongoing workflows.
  • Running the default setup requires an OpenAI API key. Other providers and local models require additional configuration.
  • Generated code is executed by the tool, so review generated changes and execution behavior before using it on important projects.
  • The README lists Python 3.10 through 3.12 as supported; Python 3.8 and 3.9 were supported only through version 0.2.6.
  • The project is positioned for experimentation. The README recommends aider for users seeking a well-maintained, hackable CLI.

Found this useful?

Share it with someone who would like gpt-engineer.

Comparisons