What is magda-agent?

Magda-agent, also known by its Russian-language title meaning "Everything on Shelves," is an experimental AI agent built in Python. It connects a Telegram chat interface to a FastAPI service that acts as the agent's core, and layers on top of that a set of components described as memory, emotion, planning, and skills. The project is described as cognitive, meaning it is exploring how to give an AI agent some persistent internal state and decision-making structure rather than just responding to individual messages. One of the more distinctive features is a self-improvement loop. The repository includes a machine-readable task queue in JSON format that lists work the agent should do to improve itself. An AI coding assistant (the README refers to one called Jules) is intended to read that queue, implement the next pending task, and update the task's status after completing it. The queue also has a policy for keeping itself replenished with new tasks. This pattern essentially treats the codebase as something an AI agent maintains and iterates on continuously. To support this, the project exposes a lightweight bridge module that a coding assistant can run from the command line without needing the full Telegram and FastAPI stack installed. The bridge can validate the task queue, report current status, identify the next task, and render a prompt. This is the intended entry point for automated improvement passes. The README is written in both English and Russian. The project is built on Python 3.12, FastAPI, and Docker, with pytest for testing. No description field was provided in the repository metadata.