Zenith Cosmo

A Python robot in the making built around an async event bus, wake-word detection, offline speech recognition, LLM-backed response generation, and text-to-speech synthesis.

Overview

Zenith Cosmo is an intelligent companion that operates entirely locally (with optional cloud LLM integrations). It features:

• Wake Word Detection: Continuous microphone monitoring using Vosk wake word recognition
• Speech Recognition: Offline transcription via Vosk STT
• Response Generation: LLM-powered responses with configurable personality and memory
• Text-to-Speech: Local synthesis using Piper TTS
• Event-Driven Architecture: Async event bus for efficient task coordination
• Local Commands: Built-in command handling for system-level operations
• Persistent Memory: SQLite-backed conversation memory and learning
• Personality System: Runtime personality controls and customization

The system is written primarily in Brazilian Portuguese and is configured through YAML settings and SQLite persistence.

Getting Started

Full for comprehensive documentation soon containig:

• System architecture and component overview
• Detailed pipeline descriptions (audio, STT, cognition, TTS)
• Event system and runtime state machine
• Configuration, logging, and data persistence
• Development workflow and testing

Quick Start

python -m cosmo.main

The entry point is cosmo/main.py, which initializes the bootstrap sequence and starts the boy.