Gemini Code - Project Story
Inspiration
Modern AI coding tools are bloated with dependencies and complex setups. We wanted a powerful coding assistant that's just one Python file—drop it anywhere and start coding.
What it does
A terminal-based AI coding assistant in ~500 lines with zero dependencies:
- Read, write, and edit files
- Search code with glob/grep
- Execute shell commands
- Markdown-formatted responses
- Works with OpenRouter & Google AI Studio
How we built it
Pure Python standard library only: urllib.request for HTTP, json for parsing, subprocess for shell, and ANSI codes for terminal styling. One file, no installs.
Challenges we ran into
- Building HTTP clients and terminal UI without third-party libraries
- Supporting multiple AI providers with different API formats
- Managing conversation context within token limits
Accomplishments that we're proud of
- 494 lines replacing tools with dozens of dependencies
- Setup in seconds:
export API_KEY=... && python geminicode.py - Clean, auditable, hackable codebase
What we learned
Python's standard library is powerful enough for production tools. Constraints force elegant solutions. Simple tool interfaces beat bloated ones.
What's next for Gemini Code
- Session persistence
- Vim/Neovim integration
Log in or sign up for Devpost to join the conversation.