video-db/call.md

Call.md is worth a look if you live in meetings and want the notes, action items and follow-up triggers handled without a separate notetaker bot joining the call. It is a macOS-first Electron app tied to a paid-by-usage cloud service, so the decision hinges on whether you accept sending audio to VideoDB.

The problem

Before a call, you open a notes app, hope you remember to capture decisions, and afterwards spend time rewriting a transcript into something a teammate or CRM can use. Call.md changes the middle of that workflow: you start a recording from the desktop app, it captures microphone and system audio, and it separates your channel from the other participants so the live transcript reads as you vs them. When the meeting ends, it produces a three-part summary, key points and action items, and can export a Markdown record with the full transcript and the intelligence generated during the call.

What's different

The loop is the selling point rather than the transcript. During recording, the app tracks metrics, surfaces coaching nudges and AI assists, and an MCP agent can detect an information need and call relevant tools. Afterwards, meeting data can be pushed to automation platforms such as n8n, Zapier or CRMs.

That design puts transcription and AI features on a network dependency: the README states these require internet connectivity and that recording, transcription and AI inputs go to VideoDB when those features are used. Local-first here means settings, history, transcripts and generated metadata live in a local SQLite database, with credential handling via OS-backed storage described in the security section; it does not mean the analysis stays on your machine.

A second trade-off is how the app is distributed. macOS gets a hosted installer and a 2-hour recording cap, with a warning 5 minutes before the stop; paused time and system sleep do not consume the allowance. Windows x64 can record only when built from source via npm run dist:win, and Linux or Windows ARM64 builds can open the UI, MCP servers, workflows and history but reject recording because no capture binary is available. Those are product boundaries, not setup bugs.

Who should try it

Good fit: a macOS user in sales, recruiting or recurring client calls who wants live assistance and a usable recap without inviting a separate bot into the meeting, and who is comfortable with VideoDB processing the audio. The workflow-automation angle also suits someone who already routes meeting output into n8n or a CRM and wants that step automatic.

Poor fit: anyone who needs an offline notetaker, a Linux recording setup, or Windows recording without building the Electron app themselves. Teams that cannot send meeting audio to an external service should stop at the privacy model. If you only need a transcript archive, the live-agent machinery is more than the job requires.

Requirements and limitations

A VideoDB API key is mandatory even for the desktop app, and transcription plus AI features need connectivity. Microphone and screen-recording permissions must be granted before the first recording, and the app reports the first transcripts after 5-10 seconds in its troubleshooting notes. The cap is fixed in code (MAX_RECORDING_DURATION_MS) and changing it means rebuilding, so long sessions need a different tool or a modified build.

Language support is uneven: the meeting language is set in Settings and applies to the next recording, and the README concedes that unsupported languages fall back to the engine default rather than failing, with a linked issue for the behavior. Ongoing upkeep means keeping the API key valid and rotating it from Settings, watching that the bundled capture executable and native SQLite module are present after updates, and rebuilding native modules when developing. Access the project instructions for the platform matrix and troubleshooting, and the example documentation for the hosted walkthrough.

HotRepo's take

HotRepo has only one dated observation: 1,888 stars and +86 on 2026-10-27, its single day on the daily Trending board. That is a snapshot of attention, not evidence that interest is rising or fading, and the repository is young enough that persistence cannot be judged. The practical question before installing is not popularity but the VideoDB dependency: confirm what the free API tier covers and how the app behaves after the 2-hour cap on your typical call length.