Engineered as a native Windows power-tool. Not another browser wrapper.
Discover how VESPER achieves sub-220ms end-to-end voice-to-action execution while guaranteeing zero window focus loss.
Hold-to-Talk: Sub-180ms Trigger
Press and hold Right Alt (or F8 / Fn) for ≥180ms. The non-activating obsidian capsule floats at the top without stealing window focus.
Release the key to transcribe via the high-speed neural engine and paste directly at your exact cursor position across any active Windows application.
VESPER Context Protocol MCP Server
VESPER runs an embedded Model Context Protocol (MCP) server. External AI agents in Cursor, Claude Code, Windsurf, and Antigravity can directly query active desktop memory.
http://127.0.0.1:32123/mcpExposes search_user_context and get_active_screen_context with zero cloud transmission.
Double-Tap Persistent Mode
Double-tap Right Alt within 380ms to lock the HUD open for long-form dictation, meetings, or multi-step prompt drafting.
Features interactive on-screen controls for Record, Pause / Resume, Stop & Insert, and Discard.
Direct Dictation vs. Smart Copilot
Pure verbatim transcription with strict capitalization, punctuation, and code formatting. Zero hallucinations.
Contextual prompt engineering designed specifically for Claude 3.7, ChatGPT, and Cursor Composer workflows.
Autonomous Screen Copilot & Safety
Multimodal visual inspection of active windows via low-latency native screen capture.
Actions with external side effects require explicit approval: press Enter to execute or Esc to reject.
Hardware Acceleration (Jensen's Law)
Co-designed for modern AI PC silicon. Operates on dedicated NVIDIA RTX Tensor Cores, Intel Core Ultra NPUs, or AMD Ryzen AI via DirectML execution providers.
Runs quantized INT8 models under 60 MB RAM with less than 1W background power, keeping laptop thermals whisper-quiet.