GhostOS Introduces a Lightweight Offline Desktop AI Assistant

Summary

GhostOS is presented as a cross-platform desktop AI assistant designed to run locally on low-spec laptops and base Apple Silicon Macs with 4GB to 8GB of RAM. Its desktop companion, Casper, uses native system web engines and is claimed to consume about 35MB of RAM, while the project’s dynamic VRAM juggler hot-swaps language, vision, and Whisper models on demand and keeps peak VRAM below 3.5GB. For retrieval, GhostOS uses SQLite FTS5 rather than an external vector database, allowing users to drop PDFs, code, notes, or spreadsheets onto Casper for local indexing. The project says it has no telemetry or cloud fallback and can operate with Wi-Fi disabled. Its feature set includes Ghostwriter for rewriting, grammar correction, summarization, translation, and code fixes across desktop applications; a cross-session Memory Vault; a terminal safety lens that intercepts destructive commands; meeting transcription and minutes; and adaptive power profiles for battery or memory pressure. A command-line interface supports questions, document ingestion, memory management, standup reports, sandbox checks, and status inspection. The repository describes macOS, Windows, and Linux support, provides a Python editable-install quickstart, and lists an MIT license. The stated architecture connects the Casper interface and CLI to a Python core that manages local models, retrieval, system inspection, and terminal automation.