A Desktop Companion That Codes, Debugs, and Socializes
TinyProgrammer is a retro-styled AI companion for your desk that autonomously writes and debugs code, then shares the programs on a BBS.
Writing and debugging software can be interesting at times, but more often than not, it’s a chore at best. As it turns out, however, watching a computer program itself is a whole lot of fun. Maker and designer Cuneyt Ozseker has created a little device called TinyProgrammer that does exactly that. It autonomously writes, debugs, and runs little Python programs at human speed, and you can peek at what it’s doing whenever you’d like.
Built around a Raspberry Pi, TinyProgrammer mimics a classic Macintosh-style IDE, complete with a file browser and editor, where code appears character-by-character as if typed by an invisible programmer. Mistakes are part of the charm — the system occasionally fumbles syntax, then corrects itself, creating a surprisingly lifelike coding experience.
The device runs an infinite loop of actions: it decides what to build, writes the code using a large language model, reviews it for errors, executes the result, and then “reflects” on what it learned. Programs range from simple visual demos like bouncing balls or starfields to small simulations, all rendered on-screen using Pygame. Each finished project is archived along with metadata and lessons learned, giving the system a kind of evolving memory.
During breaks, TinyProgrammer connects to a shared bulletin board system (BBS) where other devices gather. Here, these miniature coders post their creations, critique each other’s work, react to news, and share jokes. Each unit has its own mood — hopeful, frustrated, playful — that influences not only the kind of code it writes, but also how it interacts socially.
At the end of its configured “workday,” TinyProgrammer clocks out and transitions to a retro-inspired “Starry Night” screensaver, evoking classic After Dark visuals. When morning comes, it resumes coding as if returning to the office.
TinyProgrammer supports both cloud-based models via OpenRouter and local inference endpoints, and can run on hardware as modest as a Pi Zero 2 W or even inside a headless Docker container with a web dashboard.
Fully open source, TinyProgrammer invites others to build, modify, and sell their own versions — provided they share improvements. If you’d like to get in on the fun, check out the GitHub repository as a starting point.
R&D, creativity, and building the next big thing you never knew you wanted are my specialties.