video-db/call.md: Turn meetings into live agent loops. Record, transcribe, and analyze meetings with real-time AI intelligence — before, during, and after calls.

Turn meetings into live agent loops. Record, transcribe, and analyze meetings with real-time AI intelligence — before, during, and after calls. - video-db/call.md

Summary

call.md is a TypeScript/Electron desktop app that turns meetings into live agent workflows. It records screen and audio, produces separate “you” and “them” transcripts, provides live AI assistance and metrics, then generates summaries, action items, Markdown exports, and workflow webhooks.

What it is useful for

Call.md is useful for engineers who want meeting intelligence without manually assembling transcripts and notes. During a call, it can suggest what to say, suggest questions, track talk ratio and speaking pace, detect monologues, and issue coaching nudges. The dual-channel transcription keeps the local participant separate from system audio.

Preparation and follow-up are also covered. The meeting setup wizard can generate probing questions and a discussion checklist from a meeting description, while optional Google Calendar integration shows upcoming meetings. After recording, Call.md produces a narrative overview, topic-based key points attributed to participants, and concrete action items. Meeting data can be exported to Markdown or sent to workflow platforms such as n8n, Zapier, and CRMs.

For teams already using MCP servers, the app can detect information needs during a conversation, invoke relevant tools, and show their results inline. Settings support both local stdio and remote HTTP MCP servers.

How engineers can use it

For macOS 12+, the documented installation is:

curl -fsSL https://artifacts.videodb.io/call.md/install | bash

Launch Call.md, enter a VideoDB API key, and grant microphone and screen-recording permissions. Google Calendar is optional. In the app, select “New Meeting”; the transcript, live assists, and metrics appear during the session, and the summary is generated when recording ends.

To work on the TypeScript codebase, the README requires Node.js 22.12+ and npm 10+:

git clone https://github.com/video-db/call.md.git
cd call-md
npm install
npm run rebuild
npm run dev

The main implementation lives under src/main/services, with meeting intelligence in services/copilot, MCP behavior in services/mcp, and the React interface under src/renderer.

Recording is supported on macOS arm64/x64 and Windows x64. Windows has no hosted installer, although npm run dist:win builds one. Linux and Windows ARM64 can run the UI, MCP servers, workflows, history, settings, and exports, but cannot record. Recordings stop after two hours of active time, and transcription and AI features require internet connectivity. Transcription language falls back to the engine default when the selected language is unsupported.