Reflector: Infrastructure for Meetings, by Greyhaven

Most organizations run on meetings.

Strategy gets debated there. Decisions get made there. Context lives there. And yet, somehow, meetings are where information goes to disappear.

The Problem With Meeting Tools

Most meeting software solves the wrong problems. They record you. They help you help schedule meetings. Some even transcribe what was said. The “better” ones now provide AI summaries that compress conversations into bullet points and/or action items.

What they don’t do is preserve understanding. They don’t retain reasoning. And they certainly don’t help teams build on what happened last week, last month, or last quarter.

We felt this pain ourselves. So we built Reflector.

What Reflector Is (and Isn’t)

Reflector is not a chatbot, note-taking assistant, or transcription layer wrapped in marketing.

Reflector is an open-source system for turning live conversations into usable knowledge.

It processes meetings in real time, extracting structure as people speak. Through live processing: topics emerge naturally, speakers are identified, and summaries form as decisions are made (not hours later).

The output is designed to be edited, reused, and trusted.

Why Control and Transparency Matter

Most AI meeting tools are black boxes. Retention policies are vague and training use is unknown.

Reflector is different by design.

It is self-hosted. It is open-source. It runs where you choose, on infrastructure you control, keeping your data where it belongs.

How does it work?

Reflector’s repository is freely available on GitHub for anyone to inspect. At its core, the system is deliberately straightforward:

• Meetings use Daily.co for real-time, room-based audio and video capture, integrated into the private processing pipeline
• Audio is streamed using WebRTC and processed via WebSockets for low-latency handling
• Transcription uses open-source Parakeet or Faster Whisper
• Translation is handled by SeamlessM4T
• Topic segmentation, summarization, and analysis use open-weight LLMs such Phi-4 or GLM-4.5-Air
• All outputs are structured through an API and stored locally for editing, revision, and reuse

This architecture enables deployment on private infrastructure or controlled cloud environments without routing data through third-party SaaS platforms.

What Changes When Meetings Stop Being Disposable

When conversations stop disappearing, work changes.

Teams stop repeating the same conversations. Context stops living in people’s heads. New hires onboard faster, and meetings get better as a result.

Reflector doesn’t replace people. It helps them keep track of what matters.

Why Reflector Exists

At Greyhaven, we build AI systems for environments where trust and control matter more than novelty or convenience. Reflector is an expression of that belief.