Blog
Announcement
Introducing Raven, Your Open-Source AI Meeting Copilot
We're launching Raven, a free and open-source desktop app that transcribes your meetings in real time, cancels echo, and gives you AI assistance - all in an overlay invisible to screen sharing. Here's what it does, how it works, and why we built it.

Chaitanya Laxman
Product
Mar 4, 2026

Every day, millions of people sit through meetings where critical details slip through the cracks. You're trying to listen, think of the right response, take notes, and remember action items - all at the same time. Something always gets lost.
We built Raven to fix that.
Raven is an open-source, AI-powered meeting copilot that runs entirely on your desktop. It captures both sides of any conversation in real time, cancels echo so the remote speaker doesn't bleed into your mic, transcribes everything via Deepgram Nova-3, and gives you AI assistance from Claude or OpenAI - all in a stealth overlay that's completely invisible to Zoom, Meet, Teams, and Discord screen sharing.
Why another meeting tool?
We looked at what existed and didn't love any of it. Cloud-based meeting recorders send your audio to remote servers - a non-starter for sensitive conversations. Bot-based notetakers join your call as a visible participant, which is awkward at best and a dealbreaker at worst. Browser extensions are fragile and limited to one platform. And manual note-taking means you're not fully present in the conversation.
Raven takes a fundamentally different approach.
How it works
When you start a recording, a native binary captures system audio (the remote speaker) and your microphone simultaneously. On macOS, this uses ScreenCaptureKit for system audio and CoreAudio for the mic. On Windows, it's a Rust module using WASAPI loopback and standard capture.
Both audio streams are fed into a GStreamer pipeline running the WebRTC AEC3 engine - the same echo cancellation technology used in Google Chrome. This ensures the remote speaker's voice doesn't contaminate your mic signal. The result is two clean, separated audio streams.
These streams are sent over two parallel WebSocket connections to Deepgram Nova-3 for real-time transcription. Transcripts appear instantly in the overlay, labeled by who said what - "mic" for you, "system" for the other person.
At any point during the conversation, you can press Cmd+Enter (or Ctrl+Enter on Windows) to ask the AI for help. The AI has the full conversation context - everything that's been said by both sides - plus any documents you've uploaded to the active mode. It can suggest responses, identify follow-up questions, recap what's been discussed, or answer any custom question you type.
The stealth overlay
The overlay sits on top of your meeting window, showing the live transcript, AI suggestions, and quick action buttons. But here's the key feature: when stealth mode is enabled (it's on by default), the overlay is completely invisible to screen capture.
Raven uses Electron's setContentProtection API, which tells the operating system to exclude the window from all screen recording. On macOS, this sets the window's sharing type to none. On Windows, it uses SetWindowDisplayAffinity with the WDA_EXCLUDEFROMCAPTURE flag. The result is that when you share your screen in a meeting, your participants see your normal desktop - the Raven overlay simply doesn't exist as far as screen capture is concerned.
Stealth mode also hides the tray icon and the dock icon on macOS, leaving zero visual trace that Raven is running.
Local-first by design
Your API keys are stored locally on your machine. Your transcripts are stored in a local SQLite database. Your uploaded documents are embedded and searched locally. No audio, transcripts, or conversation data ever leaves your device unless you explicitly choose to use the optional Pro sync feature.
This isn't just a privacy promise - it's an architectural decision that you can verify by reading the source code. Raven is MIT licensed and the entire codebase is on GitHub.
What you can do with Raven today
Real-time transcription of both sides of any conversation
Echo cancellation using the WebRTC AEC3 engine
AI assistance from Claude or OpenAI with full conversation context
A stealth overlay invisible to all screen sharing apps
Customizable modes with different AI system prompts for different meeting types
Local document RAG - upload PDFs, docs, or text files and the AI references them during calls
Auto-saved sessions with complete transcripts, AI responses, and generated summaries
Keyboard shortcuts for hands-free control
Support for macOS 12+ and Windows 10/11
What's next
Raven is at version 0.1.0. It works, it's stable, and we use it daily. But there's a lot more we want to build: Linux support, more AI provider options, improved overlay customization, collaborative features for teams, and deeper integration with calendar and CRM tools.
We're building in the open and we'd love your help. Star the repo, file issues, submit PRs, or just download it and use it in your next meeting.
Your meetings are important. You should have a copilot that respects your privacy and actually helps. That's Raven.

Chaitanya Laxman
Product
Share


