What it does Natively is a desktop app for macOS and Windows that sits invisibly over your screen during video calls. It transcribes audio in real-time, analyzes screenshots, and feeds context to whatever LLM you bring (OpenAI, Claude, Gemini, or local Ollama). The pitch is two-faced: interview “copilot” on one hand, legitimate meeting assistant on the other. Your data stays local; your API keys, your bill.

The interesting bit The stealth engineering is where the craft shows. Rust-based audio capture with zero-copy ABI transfers, dual pipelines for system audio vs. microphone input, process disguising as “Terminal” or “Settings,” and dock hiding. It’s built to evade proctoring software and screen-share detection — a cat-and-mouse infrastructure applied to… job interviews.

Key highlights

• <500ms latency claimed for transcription-to-LLM roundtrips, vs. 5–90s cited for Cluely
• Local RAG memory via SQLite + sqlite-vec for querying past meetings (“What did John say about the API?”)
• Fully offline mode with Ollama, though “limited anonymous telemetry” still exists
• AGPL-3.0 license — genuinely open source, not a freemium wrapper
• BYOK (bring your own keys) — no vendor lock-in, no subscription, but also no included AI

Caveats

• The README’s user testimonials include unverifiable “private email feedback” and a premium upsell quote
• “Undetectable stealth mode” is an arms race claim; no independent security audit is mentioned
• The project is explicitly positioned as a “cheat tool” in its GitHub topics and SEO metadata, which may deter legitimate professional use

Verdict If you’re a developer who wants a hackable, privacy-respecting meeting transcription tool with local memory, this is unusually capable. If you’re looking for an interview cheating tool, it exists — but the moral hazard and detection risk are yours alone. Recruiters and hiring managers should probably know this repo has 1,362 stars and counting.