natively-cluely-ai-assistant: a stealthy real-time interview copilot with on-device STT and local RAG memory

What it solves

Natively is a real-time AI interview copilot and meeting assistant designed to provide instant, undetectable assistance during high-stakes live conversations. It eliminates the need for expensive monthly subscriptions to similar tools while ensuring user privacy by storing data locally rather than on cloud servers.

How it works

The system uses Rust-based native audio capture with Zero-Copy ABI transfers to achieve ultra-low latency (<500ms). It employs on-device speech-to-text (STT) using optimized ONNX models (such as Whisper and Moonshine) with hardware acceleration on macOS and Windows. The tool captures both system audio and microphone input via dual-channel intelligence, analyzes screen content via OCR, and feeds this context into a chosen LLM (via API keys or local Ollama models) to generate real-time responses in a stealthy overlay.

Who it’s for

Job seekers preparing for technical, behavioral, or system design interviews, as well as professionals needing a real-time assistant for sales, recruiting, or general team meetings.

Highlights

• Privacy-First Architecture: Stores transcripts, API keys, and screenshots locally by default.
• Stealth Mode: Undetectable by proctoring software; hides from the dock and disguises process names.
• Local AI Capabilities: Supports 100% offline mode using local Ollama models and on-device Whisper STT.
• Local RAG Memory: Uses SQLite vector search to allow users to query their own meeting history.
• Flexible LLM Integration: Bring-your-own-key (BYOK) support for GPT, Claude, Gemini, and Groq.
• Custom Personas: Seven tailored modes (e.g., Technical Interview, Sales) with specific system prompts and templates.