M.I.R.A. — Modular Interactive Robotic Agent

Software-first embodied AI assistant built with Python, PySide6, modular cognition, expressive state feedback, and safe local actions.

Status: Active desktop embodied assistant prototype

M.I.R.A. is a desktop embodied assistant prototype. It is software-first, not a hardware robot yet, and is designed as the foundation for future robotic embodiment.

The current prototype combines a chat/interaction core, animated eyes, a state manager, cognition modules, session memory, and safe local action execution. The goal is to build the interaction and behavior layer before moving toward physical sensors, actuators, and a robotic body.

M.I.R.A. animated desktop assistant interface
M.I.R.A. desktop interface with expressive animated eyes and compact interaction controls.

GitHub Repository

Technical Stack

Python 3.12+ PySide6 / Qt requests pytest optional Ollama integration event-driven architecture

Architecture

M.I.R.A. is organized around an event-driven interaction loop: user input flows into the chat/interaction layer, through the event bus, into the brain and intent engine, then into memory, the action executor, and response builder. The resulting state is propagated through the state manager and rendered by the face UI.

Pipeline: user input → chat/interaction layer → event bus → brain → intent engine → memory/action executor/response builder → state manager → face UI.
M.I.R.A. software architecture User input is processed by the interaction layer, event bus, brain, intent engine, memory, action executor, response builder, state manager, and face UI. User Input chat events Interaction chat layer Event Bus signals Brain cognition Intent Engine rules / local LLM Memory session context Actions safe executor State Manager assistant state Face UI eyes

Key Features

Expressive Animated Eyes

Blink, idle motion, gaze, and emotion states provide visible feedback for assistant state.

Intent Engine

Rule-based parsing handles deterministic commands, with an optional local LLM path for richer intent extraction.

Session Memory

The assistant stores recent interaction context for continuity within a session.

Safe Local Actions

A registry/executor pattern constrains local desktop actions behind explicit action contracts.

Compact GUI

The interface keeps the assistant face and chat compact, with a hidden debug drawer for development controls.

Async LLM Contract

The LLM integration path is designed to avoid UI-thread side effects and keep cognition separated from rendering.

Validation / Tests

The project includes tests covering action contracts, executor behavior, async brain contracts, desktop actions, LLM intent handling, response building, and rule-based intent parsing.

Scope

M.I.R.A. is not a hardware robot yet. The current milestone is a desktop embodied assistant prototype and a foundation for future robotic embodiment.

Source Code

The M.I.R.A. software prototype is publicly available on GitHub.

View Repository