OpenInterview

English · 繁體中文 · 简体中文 · 日本語

Local, low-latency speech recognition and an AI practice coach for job interviews on macOS (Apple Silicon). It transcribes both sides of a call on-device, spots the interviewer's questions, and shows a short answer you can say out loud, tailored to your CV and the job you're interviewing for.

Built for interview practice, mock interviews and meetings where recording and AI assistance are permitted. It is a normal, visible window with a recording indicator: no hidden overlays, no concealed recording, no screen-share evasion.

Features

• On-device speech recognition: MLX Whisper, NVIDIA Parakeet or Apple's SpeechAnalyzer, with live partials and about 1–2 s final latency. No audio leaves the Mac.
• Both sides of the call: your mic and system audio (Core Audio process tap, or BlackHole), with speaker-echo suppression.
• Question detection: technical and behavioural questions are picked out, and small talk is ignored.
• AI answers via OpenCode (e.g. DeepSeek) or any OpenAI-compatible server. Each answer is a short spoken answer, key points, and code when useful.
• Your profile: import a CV and the AI organises it. Behavioural answers use only real facts, never placeholders. A 💡 hint shows what's missing, with Add to profile.
• Per-interview target: company, role and job description shape the answers. A pre-interview check lists likely questions your profile can't answer yet.
• Languages: English, Japanese, Mandarin (Simplified / Traditional) and Cantonese.
• Session history: each session can be saved locally (transcript, Q&A, audio), played back and deleted. Saving can be turned off.
• Mock interview test: a scripted interviewer voice checks the whole chain end to end.

Quick start

Requirements: an Apple Silicon Mac on macOS 14.2+, plus Xcode or the Command Line Tools. The Apple recognizer needs macOS 26.

git clone https://github.com/kuchris/openinterview.git
cd openinterview
scripts/setup.sh            # dependencies, Swift helpers, models, environment check
uv run openinterview app    # open the app

Optional: for AI answers, run OpenCode in an empty folder, then turn on Answers (✦) in the app.

opencode auth login                         # once
mkdir -p ~/opencode-scratch && cd ~/opencode-scratch && opencode serve --port 4096

The first capture of system audio asks for macOS's System Audio Recording permission. See Installation.

Common commands

uv run openinterview app                       # the app window
uv run openinterview run --source system       # live transcript in the terminal
uv run openinterview -l yue run                # Cantonese (also ja, zh, zh-TW)
uv run openinterview profile import ~/cv.pdf   # build your profile from a CV
uv run openinterview mock --app                # scripted mock interview test
uv run openinterview check --probe             # environment and permission report

Documentation

Installation	Requirements, setup, macOS permissions
Usage	Command line, app window, sessions and history, web dashboard
Practice assistant	AI answers, your profile and CV, OpenCode setup
Languages	Japanese, Mandarin, Cantonese: models and detection
Architecture	Pipeline, system audio backends, echo handling
Benchmarks	Latency, accuracy and memory per ASR backend
Testing	Mock interview test and unit tests
PLAN.md	Original design and phase plan

Privacy

• Audio is transcribed on the Mac and never uploaded.
• Saved sessions live in ~/.openinterview/sessions/. Audio recording is a separate switch, and both can be turned off.
• Your profile (~/.openinterview/profile.md) is stored outside the repository.
• Sent to your AI provider: detected questions, recent transcript lines, your profile and the interview target. Nothing is sent while the assistant is off.
• Consent: recording the other side of a call may require their consent. Only use it where that's allowed.