InterviewCopilot

A free, open-source, real-time AI conversation assistant. It listens to live interviews, sales calls or any conversation where freezing up isn't an option, and suggests answers within seconds. It also has a mock-interview practice mode that asks you questions, then scores your answers and gives feedback.

Zero cost by default. Everything runs on your own machine: speech-to-text with Whisper and answers from a local model through Ollama. No account with any AI company is needed, and nothing leaves your computer unless you turn on a cloud provider yourself.

Use it responsibly. Some employers, assessment platforms and jurisdictions don't allow AI assistance or call transcription without everyone's consent. Check the rules for each conversation you use this in. InterviewCopilot always asks before it starts listening (unless you change that setting).

---

Features

• Live suggestions: transcribes the conversation, detects each question, works out its type, and streams a suggested answer.
• Question types, auto-detected: behavioral, "how do you…", situational, technical knowledge, system design, coding, tell me about yourself, experience. Pin a type before a session; auto-detect takes over if the conversation shifts.
• Answer style: short, balanced or long, using STAR, SMART or no framework.
• Your documents: attach up to 10 per session (resume, job description, sales script, product docs…), and answers are grounded in them.
• Session prompt: tell the AI who you're talking to and how to answer.
• Regenerate / Shorter / Longer buttons on every answer.
• Overlay: a semi-transparent, always-on-top window (opacity slider, default 85%) on Windows and macOS, and optionally inside the meeting tab via the browser extension.
• Practice mode: the AI interviews you from the job description, then scores and critiques your answers.
• History: sessions and transcripts saved locally; export any transcript to PDF.

How it works

 meeting audio ─┐
                ├─► Desktop app (Electron) ──► local backend (.NET, 127.0.0.1:5480)
 your mic ──────┘        │  overlay                ├─ Whisper (speech-to-text)
                         │                         ├─ question detection + type classification
 browser extension ◄─────┘  (optional overlay)     ├─ Ollama / your chosen model (answers)
                                                   └─ SQLite (sessions, transcripts, documents)

Mode	What it is
A: Desktop	The app notices a meeting starting, asks whether to start, and shows answers in a desktop overlay.
B: Desktop + extension	Same as A, plus answers appear inside the meeting tab in Chrome, Edge or Firefox.
C: Browser-only	Planned. A hosted version for people who don't want the desktop app (needs cloud models, so not free).

Requirements

	Minimum
OS	Windows 10 (22H2) / 11, or macOS 13 Ventura or newer
RAM	16 GB (8 GB works, slowly)
CPU/GPU	Any CPU from roughly the last 5 years. No GPU needed. Apple Silicon is fastest.
Disk	~5 GB free for models

Target speed: the first words of an answer appear 2–5 seconds after a question ends, with the default local models on the hardware above.

Getting started

Option 1: Install a release

Download the installer for your OS from Releases (-beta tags are pre-releases).

The installers are not code-signed, because signing costs money and this project stays free. Your OS will warn you the first time:

• Windows: SmartScreen shows "Windows protected your PC". Click More info → Run anyway.
• macOS: right-click (or Control-click) the app → Open → Open. On macOS 15+, if that isn't offered, go to System Settings → Privacy & Security and click Open Anyway.

Then install Ollama and pull the default model:

ollama pull llama3.2:3b

The Whisper speech model is downloaded by the app on first run.

Option 2: Run from source

Prerequisites: Node.js 24+, pnpm 10+, .NET 10 SDK, Ollama.

git clone https://github.com/ukez15/interview-copilot.git
cd interview-copilot
pnpm install
ollama pull llama3.2:3b
pnpm dev:backend     # terminal 1: backend on http://127.0.0.1:5480
pnpm dev:desktop     # terminal 2: desktop app

For the full dev stack with the Aspire dashboard, run pnpm dev instead of dev:backend.

Browser extension (optional, Mode B)

pnpm --filter @interview-copilot/extension build

• Chrome / Edge: open chrome://extensions (or edge://extensions), enable Developer mode, click Load unpacked, and select src/extension/dist/chromium.
• Firefox: open about:debugging#/runtime/this-firefox, click Load Temporary Add-on, and pick src/extension/dist/firefox/manifest.json.

The extension only works while the desktop app is running.

Choosing providers

Speech-to-text

Provider	Cost	Status
Whisper (local)	Free	Default
Deepgram	Paid, with free signup credit	Built in, disabled by default

Deepgram gives lower latency and better speaker separation. To turn it on:

1. Create a Deepgram account and API key.
2. In src/backend/src/InterviewCopilot.Api/appsettings.json, set "Transcription": { "Providers": { "Deepgram": { "Enabled": true } } }.
3. Restart the app, open Settings → Providers, choose Deepgram, and paste your key. It's stored in your OS keychain, never in a file.

AI model

Ollama (local, free) is the default. You can switch models in Settings → Providers: any model you've pulled into Ollama, or a cloud model (Anthropic, OpenAI, Google) with your own API key. Cloud usage is billed to you by that provider.

Privacy

• Audio, transcripts and documents stay on your machine by default.
• API keys are stored in Windows Credential Manager / macOS Keychain.
• Data lives in your user data folder (%APPDATA%\InterviewCopilot or ~/Library/Application Support/InterviewCopilot). Delete it to remove everything.

Contributing

Issues and PRs are welcome. Read CLAUDE.md (project rules) and GOVERNANCE.md (quality gates) first. Use pnpm, conventional commits, and keep test coverage at or above 75%.

License

MIT