Westartedbyaskingwhether“undetectable”AIwasactuallyundetectable.
Honrly started with a simple experiment: could supposedly invisible AI interview assistants actually be detected? That project became hiring software.
The open-source beginning
We built and open-sourced a local-first desktop application for macOS and Windows. It watched the machine running the interview for signals from tools such as Cluely, Interview Coder, LockedIn, and Parakeet-style assistants. The point was to make the claim testable.
The original project remains public at https://github.com/niravrohra/Honrly-app.
What that experiment could and could not show
The desktop app looked for traces on the computer in front of the candidate. That was enough to show that “undetectable” often meant “not looking.” It was not a promise that a renamed tool, a recompiled binary, or a second device would always be caught.
From detection to hiring software
Interviews were one example of a larger problem. Important decisions depend on signals whose authenticity is hard to verify. Honrly grew upstream, into the hiring system itself: post a role, send questions written for that job, and read the answer before a call.
Today, Honrly is applicant tracking with job-specific written and video assessments. The score, the quote, and the full answer live on the same record.
What the product records now
On a written answer, Honrly notes paste, autofill, and whether the question itself was copied. Those notes sit beside the response. A person still reads the work and decides. The desktop scan is the history. It is not the product you run today.
No. The desktop app is where Honrly started. The product today is hiring software: applicant tracking, job-specific assessments, and the evidence around each answer.
The macOS and Windows app is public at https://github.com/niravrohra/Honrly-app.
The next candidate is going to apply anyway.
See the work.
Write the job. Send the assessment. Read the answers before you interview.