Why We Made Raven Open Source

Raven sits in your most important conversations. Sales negotiations. Hiring decisions. Strategy discussions. Performance reviews. Board meetings. Customer calls where you're discussing pricing, timelines, and commitments.

Any tool with that level of access to your professional life needs to earn deep trust. We believe the only way to do that is full transparency. That's why Raven is open source.

Trust requires transparency

When you install a closed-source meeting recorder, you're trusting a company's privacy policy — a legal document that can change at any time. You're trusting that their servers are secure, that their employees can't access your data, and that they won't sell your conversation data to train AI models.

When you install Raven, you can read the code. You can verify that audio capture happens locally. You can see that transcription goes directly to Deepgram using your own API key with no intermediary. You can confirm that AI calls go directly to Anthropic or OpenAI using your own key. You can check that session data is stored in a local SQLite file and nowhere else.

You don't have to trust our privacy policy. You can trust the code.

Privacy should be verifiable, not promised

Raven's architecture is local-first by design. Here's what that means in concrete terms:

• Audio capture - runs as a native process on your machine. Raw audio never leaves your device; it's processed locally through the GStreamer echo cancellation pipeline.

• Transcription - the clean audio is sent to Deepgram via WebSocket using your API key. Raven has no server in between.

• AI assistance - prompts are sent directly to Anthropic or OpenAI using your API key. Again, no intermediary server.

• Document embedding - documents are parsed, chunked, and embedded entirely locally using @xenova/transformers. The embedding model runs in your Electron process.

• Session storage - transcripts, AI responses, and summaries are stored in a local SQLite database at ~/Library/Application Support/project-raven/ (macOS) or %APPDATA%/project-raven (Windows).

• No telemetry - Raven doesn't phone home. There's no analytics, no usage tracking, no crash reporting. You can verify this by searching the codebase.

Every one of these claims is verifiable by reading the source code. We're not asking you to take our word for it.

Meeting tools should be extensible

Meetings are wildly varied. A venture capitalist uses meetings differently than a recruiter, who uses them differently than a support engineer, who uses them differently than a product manager. No single set of features will serve everyone perfectly.

With the source code available, you can:

• Add new AI providers - the provider system uses a factory pattern that's designed to be extensible

• Modify the overlay UI - it's React with Tailwind, so customization is straightforward

• Extend the session storage - add custom fields, build integrations with your CRM or project management tool

• Create custom quick actions - beyond the built-in assist/follow-up/recap actions

• Build plugins - the Electron architecture supports IPC-based extensions

We've seen people fork Raven to add features we never imagined. That's exactly what we hoped would happen.

Open source drives better software

We want Raven to be the best meeting copilot available. Being open source helps us get there faster for several reasons:

Bug discovery. Users who can read the code find and report bugs with much more context - often with a pull request attached. "The AEC pipeline breaks when the mic sample rate is 44.1kHz but the system audio is 48kHz" is infinitely more useful than "transcription is sometimes garbled."

Security. Meeting copilots are high-value targets for security researchers. Having the source code public means security-conscious users and researchers can audit the codebase, find vulnerabilities, and help us fix them before they become problems.

Trust. Open source software generates more trust, which drives more adoption, which generates more feedback, which makes the software better. It's a virtuous cycle.

Diverse use cases. We build Raven for our own meetings, which means we have blind spots. Contributors who use Raven in contexts we haven't encountered — conference talks, therapy sessions, legal depositions, language learning - push the product in directions we wouldn't have explored on our own.

How we sustain development

A reasonable question: if Raven is free and open source, how do you make money?

Raven offers a Pro tier with features that benefit from a server component:

• Managed API keys - no need to bring your own Deepgram, Anthropic, or OpenAI keys

• Cloud sync - access your sessions across devices

• Team features - shared modes, team session libraries, admin controls

The core meeting copilot - audio capture, echo cancellation, transcription, AI assistance, stealth overlay, modes, RAG, sessions - is and will remain free and open source. The Pro features are for users and teams who want a managed, zero-configuration experience.

This model aligns our incentives with our users. The open-source product has to be genuinely good on its own, because that's what drives Pro adoption. We can't hold essential features hostage behind a paywall - the community would fork around them.

Get involved

Raven is MIT licensed. Clone the repo, file issues, submit pull requests, or just use it and tell us what you think.

The codebase is TypeScript (Electron + React + Vite), with native modules in Swift (macOS) and Rust (Windows). If you're interested in audio processing, desktop apps, AI integrations, or meeting productivity, there's something here for you.

Star us on GitHub, and join us in building the meeting copilot that everyone deserves.