Call.md
Turn meetings into live agent loops. Record, transcribe, and analyze meetings with real-time AI intelligence — before, during, and after calls.
Explore the docs »
View Demo
·
Install
·
Report Bug
Demo
https://github.com/user-attachments/assets/94470e99-c0f6-4e35-9d03-b28efa362b3b
Quick Install
macOS (Apple Silicon & Intel):
curl -fsSL https://artifacts.videodb.io/call.md/install | bash
After installation:
- Launch Call.md from Applications or Spotlight
- Grant system permissions when prompted, or configure them later from Settings
- Register with your VideoDB API key (get one free)
Microphone and screen-recording permissions are required before the first
recording. Google Calendar is optional and can be connected or skipped during
onboarding.
| Platform |
Installer |
Status |
| macOS 12+ (Apple Silicon & Intel) |
curl command above |
Supported |
| Windows x64 |
Build from source with npm run dist:win |
Recording supported; no hosted installer yet |
| Windows ARM64 |
— |
Recording not supported |
| Linux |
Build from source with npm run dist:linux |
App features available; recording not supported |
The VideoDB capture SDK now ships recording binaries for darwin-arm64,
darwin-x64, and win32-x64. Call.md verifies that the capture executable and
SQLite native module are present in the packaged app.
Windows x64 recording is supported by the source build, but the project does
not currently publish a Windows installer. Linux and Windows ARM64 builds can
run the UI, MCP servers, workflows, history, settings, and exports, but the app
will reject recording before launch because no capture binary is available.
See Building for other platforms.
Overview
Call.md turns meetings into live agent loops. It records locally, transcribes in real-time (you vs them), and provides live intelligence during calls. When the meeting ends, it generates summaries with action items and can send data to your workflow automation platforms.
Features
During the Meeting (Live Intelligence)
- Dual-Channel Transcription - Separate transcription for you (mic) vs them (system audio), powered by VideoDB
- Transcription Language - Pick the meeting language in Settings → Transcription, or leave it on Automatic
- Live Assist - AI generates contextual suggestions: things to say, questions to ask
- Conversation Metrics - Real-time monitoring of talk ratio, speaking pace (WPM), questions asked, monologue detection
- Coaching Nudges - Gentle rate-limited alerts when conversation needs steering
- MCP Auto-Triggering - Detects information needs from conversation and calls your MCP tools automatically
- MCP Results Panel - Inline display of tool outputs (markdown, links, structured data) during meetings
- Bookmarking - Mark important moments for easy reference later
Post-Meeting Intelligence
- AI-Generated Summaries - Three parallel extractions:
- Short overview (narrative summary)
- Key points by topic (attributed to participants)
- Action items (concrete next steps)
- Structured Export - Markdown export with full transcript, summary, and metrics
- Workflow Webhooks - Auto-send meeting data to n8n, Zapier, or CRMs when meeting ends
Meeting Preparation
- Meeting Setup Wizard - AI-generated probing questions based on meeting description
- Dynamic Checklist - AI creates discussion checklist from meeting context
- Google Calendar Integration - Sync upcoming meetings
Privacy & Storage
- 2 Hour Recording Limit - Recordings stop themselves after 2 hours of active recording time, with a warning 5 minutes before; pauses and system sleep do not consume the allowance
- Local-First - Settings, meeting history, transcripts, and generated metadata are stored in the local SQLite database
- Screen & Audio Recording - Capture screen, microphone, and system audio simultaneously
- Recording History - Browse and review past recordings with full transcripts
- VideoDB Integration - Transcription and AI features require internet connectivity
- Account Controls - Validate and rotate the VideoDB API key from Settings, or log out and clear persisted session and Google credentials
How It Works
During Recording:
- Captures dual-channel audio (you vs them) and sends to VideoDB for real-time transcription via WebSocket
- Runs live intelligence: metrics tracking, coaching nudges, and AI-generated assists
- MCP agent automatically detects information needs and triggers relevant tools
After Recording:
- Generates three-part summary: narrative overview, key points, and action items
- Sends meeting data to workflow automation platforms (n8n, Zapier, CRMs)
- Exports to markdown with full transcript and intelligence
Tech Stack
- Electron 42 - Desktop application framework
- TypeScript 5.8 - Full type safety across main and renderer processes
- React 19 - Modern UI framework with concurrent features
- Tailwind CSS + shadcn/ui - Utility-first styling with high-quality component primitives
- tRPC 11 - End-to-end type-safe API layer between main and renderer
- Hono - Fast HTTP server for tRPC API endpoints
- Drizzle ORM + SQLite - Type-safe database operations with local storage
- Zustand - Lightweight state management
- VideoDB SDK (0.3.0) - Screen recording, transcription, and video processing
- MCP SDK (1.0.0) - Model Context Protocol for tool integrations
- OpenAI SDK (6.19.0) - LLM calls via VideoDB's OpenAI-compatible API
- Vite - Fast frontend bundling and hot module replacement
Prerequisites
- macOS 12+ (Monterey or later) or Windows x64 — required for recording, see Platform Support
- VideoDB API Key (console.videodb.io)
- System permissions: Microphone and Screen Recording
For development: Node.js 22.12+ and npm 10+
Getting Started (Users)
Install:
curl -fsSL https://artifacts.videodb.io/call.md/install | bash
Launch the app and enter your VideoDB API key (get one free)
Grant permissions when prompted, or configure them later in Settings
Start Recording - Click "New Meeting" and begin your first session
The app will transcribe in real-time, show live assists, and generate a summary when you're done.
Getting Started (Developers)
Clone the repository:
git clone https://github.com/video-db/call.md.git
cd call-md
Install dependencies:
npm install
Rebuild native modules for Electron:
npm run rebuild
Start development mode:
npm run dev
Register with your VideoDB API key when the app opens
Available Scripts
| Command |
Description |
npm run dev |
Start development mode (main + renderer with hot reload) |
npm run build |
Build TypeScript and React for production |
npm run dist:mac |
Build macOS distributable DMG |
npm run dist:win |
Build Windows x64 NSIS installer with recording support |
npm run dist:linux |
Build Linux AppImage (recording unavailable, see below) |
npm run typecheck |
Run TypeScript type checking |
npm run test |
Run unit tests |
npm run lint |
Run ESLint |
npm run rebuild |
Rebuild native modules for Electron |
npm run db:generate |
Generate database migration files |
npm run db:migrate |
Apply database migrations |
electron-builder is already configured for Windows (NSIS) and Linux (AppImage),
so you can produce an installer on those platforms:
npm ci
npm run dist:win
Build and run the release candidate on its target OS. The project uses published
target prebuilds for better-sqlite3 and packaged VideoDB capture binaries, so
a macOS development machine can cross-package a Windows x64 directory build for
structural verification. That does not replace native Windows testing or
installer signing. This follows
electron-builder's multi-platform build guidance,
which requires target prebuilds for native dependencies and still needs
target-platform validation.
Recording support is available on macOS arm64/x64 and Windows x64. On Linux and
Windows ARM64, the full UI, MCP servers, workflows, settings, history, and
markdown export remain available; starting a recording returns a clear
unsupported-platform error.
Release artifacts must be signed with the project's platform credentials.
macOS releases also need notarization and stapling. Unsigned local builds can
receive a different Keychain or OS-permission identity on every rebuild and are
not representative of the installed release.
MCP Server Setup
Connect MCP servers in Settings → MCP Servers:
- Click Add Server
- Choose transport: stdio (local) or http (remote)
- Configure and click Connect
The MCP agent runs automatically during meetings, detects information needs from conversation, and triggers relevant tools. Results appear inline in the MCP Results panel.
Development
Project Structure
src/
├── main/ # Electron Main Process
│ ├── db/ # Database layer (Drizzle + SQLite)
│ ├── ipc/ # IPC handlers
│ ├── lib/ # Utilities (logger, paths, permissions)
│ ├── server/ # HTTP server (Hono + tRPC)
│ │ └── trpc/ # tRPC router and procedures
│ └── services/ # Business logic
│ ├── copilot/ # Meeting intelligence services
│ │ ├── context-manager.service.ts
│ │ ├── conversation-metrics.service.ts
│ │ ├── nudge-engine.service.ts
│ │ ├── sales-copilot.service.ts # Core orchestrator
│ │ ├── summary-generator.service.ts
│ │ └── transcript-buffer.service.ts
│ ├── mcp/ # MCP orchestration and tool execution
│ │ ├── connection-orchestrator.service.ts
│ │ ├── intent-detector.service.ts
│ │ ├── mcp-agent.service.ts
│ │ ├── tool-aggregator.service.ts
│ │ └── result-handler.service.ts
│ ├── live-assist.service.ts
│ ├── mcp-inference.service.ts
│ ├── llm.service.ts
│ └── videodb.service.ts
├── preload/ # Preload scripts (IPC bridge)
├── renderer/ # React Frontend
│ ├── api/ # tRPC client
│ ├── components/ # UI components
│ │ ├── auth/ # Authentication modal
│ │ ├── calendar/ # Calendar integration UI
│ │ ├── copilot/ # Meeting intelligence UI
│ │ ├── history/ # Recording history views
│ │ ├── home/ # Home screen
│ │ ├── icons/ # Icon components
│ │ ├── layout/ # App layout (sidebar, titlebar)
│ │ ├── mcp/ # MCP results/status components
│ │ ├── meeting-setup/ # Meeting prep wizard
│ │ ├── recording/ # Recording controls & live assist
│ │ ├── settings/ # Settings editors
│ │ ├── transcription/ # Live transcription panel
│ │ └── ui/ # shadcn/ui components
│ ├── hooks/ # Custom React hooks
│ ├── lib/ # Utilities
│ └── stores/ # Zustand state stores (session, copilot, mcp)
└── shared/ # Shared types & schemas
├── schemas/ # Zod validation schemas
└── types/ # TypeScript types
IPC API
The app exposes IPC APIs through the preload script:
window.electronAPI.mcp.* - MCP server and tool operationswindow.electronAPI.mcpOn.* - MCP event subscriptions
Permissions
The app requires the following permissions before recording:
- Microphone - For voice recording
- Screen Recording - For screen capture
On macOS, grant them in System Settings → Privacy & Security. Newer macOS
versions label the screen permission Screen & System Audio Recording. On
Windows, enable microphone access for desktop apps when prompted. You can skip
permission setup during onboarding and return to it from Settings, but recording
will remain unavailable until the required permissions are granted.
Troubleshooting
Recording not starting:
- Check microphone and screen recording permissions in System Settings
- Verify VideoDB API key is valid
- Confirm the platform is macOS arm64/x64 or Windows x64
Transcription not appearing:
- Ensure mic and system audio are enabled in settings
- Wait 5-10 seconds for first transcripts
- Check internet connectivity
A recording stopped on its own:
- Recordings are capped at 2 hours and stop automatically when they reach it.
You get a system notification 5 minutes before, and the recording is saved
and summarised exactly as if you had pressed Stop
- Paused time does not count toward the limit, so the cutoff matches the
elapsed timer shown during the meeting
- Time spent in system sleep does not count either
- To change the cap, edit
MAX_RECORDING_DURATION_MS in
src/shared/constants/recording.ts and rebuild
Transcription is in the wrong language:
- Set the meeting language in Settings → Transcription (it applies to the
next recording, not one already in progress)
- If transcripts still come back in English, the language is not yet supported
by the VideoDB transcription backend. The app sends
language_code and falls
back to the engine default rather than failing — see
#25
Development issues:
- Rebuild native modules:
npm run rebuild
- Check Node.js version (requires 22.12+)
- Review logs:
~/Library/Application Support/call-md/logs/
Data Storage
Application data is stored in:
~/Library/Application Support/call-md/
├── config.json # Settings and encrypted desktop access token
├── data/
│ └── call-md.db # SQLite database; sole encrypted API-key authority
├── google_tokens.enc # Encrypted Google OAuth tokens, when connected
└── logs/
└── app-YYYY-MM-DD.log # Daily log files
Windows stores the equivalent files under the Electron application-data
directory for the current user.
Security
The application database, settings, and logs stay on your machine. Recording,
transcription, and AI inputs are sent to VideoDB when those features are used.
If you enable Google Calendar, remote MCP servers, or workflow webhooks, the
relevant data is also sent to the services you configure.
- Credentials at rest — the encrypted SQLite user row is the sole authority
for the VideoDB API key; it is no longer duplicated in
config.json. The
desktop access token in the config is encrypted, while the database stores
only its SHA-256 digest. Google OAuth tokens use Electron
safeStorage. MCP
server environment variables and HTTP headers use AES-256-GCM under a
keychain-wrapped key. Storage uses Keychain on macOS, DPAPI on Windows, and a
strong libsecret backend on Linux; credential writes fail closed when strong
OS-backed storage is unavailable or Linux selects the insecure basic_text
backend.
- Account changes — a replacement VideoDB API key is verified before the
single database update and cannot be changed during an active meeting.
Logging out is completed in the main process: capture and calendar activity
stop, the local access token is invalidated, Google tokens are cleared, and
the renderer changes state only after persistence succeeds.
- Local API — the tRPC server binds to
127.0.0.1 only and accepts CORS
requests from loopback origins, so nothing on your network can reach it.
Every procedure except registration requires a valid access token.
- File permissions — the app data directory is
0700 and the database,
config, tokens and logs are 0600.
- Renderer — both windows run with
contextIsolation, no Node integration,
and the Chromium sandbox enabled. The API key is never written to
localStorage. These controls follow Electron's
security checklist.
- Webhooks — workflow URLs are validated at save time and at call time.
Non-HTTP(S) schemes, embedded credentials, redirects, and hosts that resolve
to loopback, private, link-local, special-use, or cloud-metadata addresses are
rejected. IPv4-mapped IPv6 addresses are classified by their embedded IPv4
destination. Delivery is pinned to the addresses approved by that DNS lookup,
preventing DNS rebinding between validation and connection, and requests time
out if the response does not finish within 30 seconds.
- Logs — credential-shaped fields are redacted before anything is written.
Upgrades migrate existing data in place on first launch; you do not need to log
in again. Report security vulnerabilities privately to
[email protected], not through a public issue.
Made with ❤️ by the VideoDB team
Call.md 把会议变成实时的 agent 循环：它实时录制、转写并分析通话，通话中给你 AI 实时辅助，会后生成带行动项的 AI 摘要。