mini-agi

Why we included this project

MiniAGI shows how a working autonomous agent can be built from just a few pieces: a solid system prompt, a minimal toolset for running shell commands and Python, and a short-term memory that summarizes as it goes. Give it an objective, like drawing a car or assembling a small website, and it plans, executes steps, and self-corrects until the task is done. Because it targets the OpenAI API and keeps its own footprint tiny, it reads well as a compact reference for how chain-of-thought, inner monologue, and self-criticism fit together in practice. Teams learning agent design will appreciate being able to trace the entire loop in one small codebase. The optional critic mode is a handy on/off switch for seeing how extra self-review changes output quality relative to API cost.