Ghostmeet

Ghostmeet is a self-hosted meeting assistant that transcribes your calls and writes up summaries without sending a single second of audio to someone else's server. It runs as an invisible Chrome side panel, captures the audio from a browser tab, and does the transcription right there on your own machine. No bot shows up in the call, no third-party service records the room, and nothing gets uploaded anywhere unless you explicitly ask for a summary. For a growing number of people, that's the difference between using a meeting-notes tool and quietly deciding not to.

The problem it solves is the low-grade discomfort a lot of teams feel about the convenient options. Most meeting-notes products work by sending a bot into your Zoom or Meet call and streaming everything to a cloud you don't control. That's fine for a throwaway standup and a genuine problem for a coaching session, a user interview under NDA, or any conversation a client would rather not see leave the building. Ghostmeet keeps the whole pipeline local, so the awkward question of where your recordings actually live mostly disappears.

It works through Chrome's tab-capture API. Audio from Meet, Zoom, Teams, Discord, or any other tab is streamed to a local FastAPI backend, transcribed by Whisper with a choice of model sizes so you can trade speed against accuracy, and turned into live captions that refresh every few seconds while the meeting is still going. The whole stack runs on localhost, and the transcription step never touches the network at all. What you get is a running transcript on your own screen, produced by software you can see and control.

Running it as an invisible side panel matters more than it sounds. Because Ghostmeet reads the tab's audio directly instead of dialing into the meeting as a participant, there's no extra attendee for anyone to notice, no recording banner, and no changed dynamic in the room. Everyone talks the way they normally would, and you still walk away with a full transcript. The choice of Whisper model size lets you tune that transcript, a smaller model for a fast, low-overhead pass on a quick call, a larger one when you need the accuracy to hold up for something you'll act on later. It's a deliberately small surface area with one firm rule, capture everything locally and send nothing out unless you press the button for a summary.

Summaries are a separate, deliberate step. When you actually want a write-up, Ghostmeet sends the transcript to the Claude API using your own key and returns a structured summary with the decisions that were made, the action items that came out of it, and the next steps. You can also feed it context before or during the call, an agenda, a design doc, or a prior transcript, so the summary understands what the meeting was really about instead of guessing from raw dialogue. Because that Claude call only fires on demand, the one privacy-sensitive and paid part of the process is something you choose rather than something that happens automatically in the background.

That split is the heart of the design. Accurate live transcription is free and completely local, and the AI summary is an occasional, opt-in extra you reach for when a particular meeting is worth it. It means a day full of quick internal calls costs you nothing, while the one client conversation you actually need documented gets a clean, structured recap for the price of a single API request. You never pay for a summary you didn't ask for, and you never wonder whether a background process is quietly shipping your audio somewhere.

It's aimed at a few clear audiences. Engineers who want tidy standup notes and a reliable list of action items. Managers and coaches running sensitive one-on-ones where a cloud recording would change the tone of the conversation. And researchers conducting user interviews they've promised, sometimes contractually, to keep private. More broadly it fits anyone who has liked the idea of automatic meeting notes but hesitated to hand a vendor a steady stream of their calls.

Ghostmeet is open source under the MIT license and free to run. Setup is a one-command Docker deployment that brings the stack up on localhost, and you supply your own Anthropic API key for the summary feature, so the only running cost is whatever those summaries consume. Issues, the roadmap, and discussion live on the project's GitHub repository. It's a younger, self-hosted tool rather than a polished commercial service with a support desk, which is exactly the trade a privacy-first team tends to be happy to make.