AI Private

AIOS

AIOS is a natural-language AI Operating System for macOS that interprets voice or text commands and executes real OS-level actions — launching apps, controlling browsers, organizing files, running terminal commands. It is a fully local, privacy-preserving alternative to Siri and Cortana with significantly broader OS control.

Contact Me for Access

The Problem

Existing voice assistants (Siri, Cortana, Google Assistant) have shallow OS integration, require cloud connectivity, and expose user data to third parties. Power users and developers needed a local-first assistant that could execute complex, multi-step OS tasks from natural language without leaving a data trail.

The Solution

A three-layer architecture: a spaCy-based intent engine (15+ action categories, 98% intent classification accuracy) handles direct commands; a RAG module with ChromaDB stores previously learned command patterns for continuous improvement; Ollama Phi-3 Mini serves as a fallback for complex or ambiguous queries that require reasoning. All processing is local.

Key Features

spaCy intent engine covering 15+ OS action categories

Voice input (SpeechRecognition) + text input

Executes: app launch, browser control, file management, terminal commands

RAG-based self-learning from new command patterns (ChromaDB)

Ollama Phi-3 Mini LLM fallback for complex queries

Electron cross-platform desktop app

React frontend with command history and action preview

100% local — no cloud calls, no data collection

Challenges & Learnings

Safely executing OS-level actions from natural language requires a robust permission model to prevent destructive commands (e.g., 'delete files' without confirmation). A two-phase execution architecture was designed: intent classification → action preview requiring user confirmation for destructive operations. The RAG module's ChromaDB vector store needed careful chunking strategy to distinguish semantically similar but operationally different commands.

Business Impact

Enables power users and developers to execute complex OS workflows via voice in a fully local, private environment. Particularly useful for accessibility (hands-free computing) and developer productivity (voice-driven terminal and IDE control).

Future Direction

Plugin architecture for third-party integrations (VS Code, Figma, Slack), multi-step workflow macros, and a fine-tuned smaller model replacing Phi-3 Mini for faster intent resolution.

Tech Stack

PythonspaCyFlaskOllamaPhi-3 MiniElectronReactLangChainChromaDB (RAG)

Quick Links

Request Access

Want to see more of my work?

Back to Projects