blog
Field reports.
From the inside.
Technical depth on what we shipped, what broke, and what we learned. No SEO-bait listicles, no thought-leadership filler.
After the Cluely data exposure — what candidates should actually be using
2026-05-17
Cluely's 2025 transcript-exposure incident underlined what every privacy-aware candidate already suspected — interview transcripts are sensitive, and most copilots treat them as marketing telemetry. Here's what we built differently, and what to look for in any tool you trust with a live interview.
How interview copilots actually work — a technical anatomy
2026-05-03
A walkthrough of the audio-capture, STT, LLM, and stealth pipeline that powers a real-time interview copilot. With code references and latency budgets.
Latency teardown — six interview copilots, end-to-end, same machine, same question
2026-05-16
We ran the same behavioral question through Mirly, Final Round AI, Cluely, Parakeet AI, LockedIn AI, and Sensei AI on the same MacBook Air M2 over the same network. End-to-end first-token times ranged from 127ms (Mirly) to 1,810ms (Final Round). The full methodology, raw numbers, and why everyone except us inflates their published claim.
Stealth by documented OS API, not by trick — how Mirly stays out of screen share
2026-05-15
Mirly's overlay is invisible to Zoom, Teams, Meet, and Webex because it uses NSWindow.sharingType on macOS and SetWindowDisplayAffinity on Windows — the same APIs 1Password and Netflix use. No kernel exploit, no hidden process, no thing-that-will-break-next-Tuesday.
Why we don't auto-renew (and why nobody else in this category can say that)
2026-05-14
Half of Final Round AI's 1-star Trustpilot reviews are billing complaints. We built Mirly's pricing as the explicit inverse — opt-in renewal, 5-day reminder email, never a silent charge. Here's why we made it the headline.