How abheytiwari/indica works
Project type and purpose
• INDICA is an open-source, Python-based voice-powered virtual assistant. It uses speech recognition (listener.py), Gemini LLM (ai_engine/gemini_engine.py), a modular action system (modules/ folder), and text-to-speech (tts.py) to “think, talk, and act.”
• Its goal is to parse voice commands, decide whether to invoke a knowledge task (LLM) or system task (apps, email, weather, etc.), maintain short- and long-term memory (logs/ and modules/LTM.py), then respond via voice.
File structure
• Top-level orchestration:
• main.py – application entry point
• listener.py – captures audio, STT
• dispatcher.py – routes intents to modules or AI engine
• tts.py – speaks replies
• config.py – API keys, settings
• AI Engine:
• ai_engine/gemini_engine.py – wraps Google Gemini API calls
• ai_engine/gemini_research.py – research-oriented prompts
• Modules folder: each file implements a discrete system action, e.g., email.py, weather.py, apps.py, memory.py, fun.py, wiki.py, etc. modules/LTM.py handles long-term memory persistence.
• Logs:
• logs/logs.txt – short-term memory of last 5 commands
• logs/LTM.json – long-term embeddings store (planned)
• research_outputs/ – stores artifact outputs from “research” tasks.
• requirements.txt – dependency list.
README insights
• User → STT → Intent Recognizer → Decision Engine → either RAG (LLM + Memory) or System Control (modules) → TTS → User
• It lists external integrations: Gemini API, weather API, email (SMTP), Wikipedia, pyjokes, webbrowser, subprocess for app launches.
Detail without overcomplicating
• Module group, the AI engine, memory stores, and the voice I/O clearly.