The open-source AI interview copilot & meeting assistant. Real-time transcription, AI-suggested answers, and screenshot reasoning — running on your machine.
Pika listens to your meetings, transcribes both sides of the conversation in under 500 ms, and quietly suggests answers powered by the LLM you choose — OpenAI, Claude, Gemini, Groq, or fully offline with Ollama. Bring your own API key, keep your audio on your machine, and customize every shortcut.
If you've been looking for an open-source alternative to Cluely, Granola, or Otter that respects your privacy and your wallet, Pika is built for you.
Full demo video: demo/out/pika-demo.mp4 · Built with Remotion.
| Pika | Cluely | Granola | Otter | |
|---|---|---|---|---|
| Open source | ✓ AGPL-3.0 | — | — | — |
| Bring your own API key | ✓ | — | — | — |
| 100% offline option (Ollama) | ✓ | — | — | — |
| Real-time AI answer suggestions | ✓ | ✓ | — | — |
| Stealth overlay (hidden in screen-share) | ✓ | ✓ | — | — |
| Local RAG memory across past meetings | ✓ | — | — | — |
| Custom global keyboard shortcuts | ✓ | ✓ | partial | — |
| Audio never leaves your device | ✓* | — | — | — |
* When using Ollama for LLM and a local STT engine. With cloud providers, audio is sent only to the provider you configured.
- Interview practice & prep — rehearse with realistic AI-suggested answers, not generic flashcards.
- Sales & customer success calls — surface objection handlers, pricing, and product facts in real time.
- Customer support — instant context from your past tickets via on-device vector search.
- Lectures & research — live transcribe, summarize, and ask follow-up questions on what you just heard.
- Accessibility — live captions for hearing-impaired users with provider-agnostic STT.
- Sub-500 ms transcription powered by a Rust native audio module (cpal + WebRTC VAD + rubato resampling).
- Dual-channel capture — system audio (the meeting) and your microphone are streamed independently for clean attribution.
- Any LLM, any STT — switch providers at runtime: OpenAI, Anthropic Claude, Google Gemini, Groq, Ollama, or any OpenAI-compatible endpoint.
- Local RAG memory — past meetings indexed with
sqlite-vec; search and recall without sending data anywhere. - Screenshot reasoning — capture full screen or a region, pipe it to a vision model, get an answer in seconds.
- Stealth overlay — disguise the process name and hide the window from screen-sharing.
- Configurable everything — every shortcut, every provider, every key in your OS keychain.
- Cross-platform — macOS (Apple Silicon + Intel) and Windows 10/11.
Grab the latest signed build from GitHub Releases.
macOS unsigned warning: If macOS reports the app as unverified, clear quarantine:
xattr -cr "/Applications/Pika.app"
git clone https://github.com/royisme/pikabaka.git
cd pikabaka
pnpm install # also rebuilds native modules for Electron ABI
pnpm run build:native # compile Rust audio module
pnpm run app:dev # Vite + Electron in dev mode
pnpm run dist # production build via electron-builderPrerequisites: Node.js 20+, pnpm, Rust toolchain (for the native audio module), Git.
You only need one LLM provider and one STT provider to get started. All keys are stored in your OS keychain.
| Provider | Best for | Notes |
|---|---|---|
| Google Gemini | Cost + huge context window | Recommended default for most users |
| OpenAI | General quality, GPT-4o, o3 | Strong at chained reasoning |
| Anthropic Claude | Coding interviews | Best Claude tier for technical depth |
| Groq | Sub-second inference, vision | Cheapest fast tier |
| Ollama | Fully offline, no key required | Run Llama/Qwen/etc. on your machine |
| Custom | Any OpenAI-compatible endpoint | Self-hosted vLLM, LM Studio, etc. |
Google Cloud Speech (default), Deepgram Nova-3, Soniox, OpenAI Whisper, Groq Whisper, ElevenLabs Scribe, Azure Speech, IBM Watson. Auto-fallback to Google on failure.
Three-process Electron app:
- Main process (
electron/) — window orchestration, IPC, credential storage, provider routing. - Renderer (
src/) — React 18 + TypeScript + Tailwind CSS + React Query. - Native module (
native-module/) — Rust (NAPI-RS) for low-latency audio capture; CoreAudio/ScreenCaptureKit on macOS, WASAPI on Windows.
┌─────────────────────────────────────┐
│ Electron Main (Node + native deps) │
│ IPC · keychain · provider routing │
└────────────┬────────────────────────┘
IPC bridge │
┌────────────┴───────────┐ ┌──────────────────┐
│ Renderer (React/Vite) │ │ Rust (NAPI-RS) │
│ Overlay · Settings · UI│ │ Audio capture │
└────────────────────────┘ └──────────────────┘
Local data lives in ~/Library/Application Support/Pika (macOS) or %APPDATA%\Pika (Windows). API keys live in the OS keychain. No telemetry runs by default.
- macOS 12+ (Apple Silicon or Intel) or Windows 10/11.
- 4 GB RAM minimum, 8 GB recommended.
- 16 GB+ if running Ollama locally.
- Linux build (Wayland audio capture)
- Plugin / extension API
- iOS companion for cross-device handoff
- More languages in the bundled embedding model
- Hosted cloud sync (opt-in, end-to-end encrypted)
Track progress in GitHub Issues.
Is Pika free? Yes — AGPL-3.0. Free for personal use, study, and open-source projects. Commercial / proprietary integrations should review the AGPL terms or reach out about a commercial license.
Does my audio leave my machine? Only if you choose a cloud STT/LLM provider. Pair Ollama with a local STT engine (or system-only transcription) for a fully offline pipeline.
Where are my API keys stored? In your OS keychain — Keychain Access on macOS, Credential Manager on Windows. Never in plaintext config files.
How is Pika different from Cluely? Open source under AGPL-3.0, no subscription, you bring your own API key, and you can run completely offline with Ollama. Every shortcut and provider is configurable.
Why "stealth mode"? The overlay can be hidden from screen-sharing tools so private notes during a sales demo don't show on the customer's screen. It's a privacy feature, not a permission to break rules — please use Pika ethically and respect any policies that apply to your conversations.
Can I use Pika during a coding interview? Use it for prep and rehearsal before the interview — that's the legitimate use case. Real interviews almost always have explicit rules about external assistance; respect them.
Does it work on Linux? Not yet. Tracking on the roadmap.
Pull requests welcome — see CONTRIBUTING.md and CODE_OF_CONDUCT.md. For security issues, see SECURITY.md.
AGPL-3.0 — see file for details.
If Pika is useful to you, give it a star — it helps others discover the project.
Keywords: AI interview copilot, AI meeting assistant, real-time transcription, open source Cluely alternative, local AI assistant, Electron AI app, BYO API key meeting tool, privacy-first AI copilot, screenshot AI assistant, RAG meeting memory.