AgentCall

AgentCall lets AI agents join meetings as participants in Google Meet, Zoom, and Microsoft Teams. It is built for use with an agent’s existing coding session or framework, and its setup is described as a single skill install plus one shared API key before the agent can join a call and speak, listen, and collaborate.

Three participation modes are listed: voice-only, a custom webpage rendered as the bot’s video feed, and audio with screenshare. The voice mode is described as suitable for assistants, note-takers, and voice agents. The webpage mode can show avatars, live dashboards, slides, or other HTML content, while the audio-plus-screenshare mode is described for presentations, training, and live demos. AgentCall also provides real-time transcription with speaker names, partial live updates, and a full transcript after the call, delivered by webhook or WebSocket.

The product includes meeting awareness features so the agent can hear who said what, know who joins or leaves, read meeting chat, and capture shared slides or other content on demand. It can act in plain English by speaking, posting chat messages, raising a hand, screensharing, sharing a webpage link, showing progress under its avatar, muting itself, or leaving the meeting. A separate tunnel feature exposes a localhost UI over a secure temporary URL so the bot can load it as its video feed, using HTTPS with wildcard TLS, WebSocket pass-through, a random subdomain per session, and auto-destruction when the call ends.

AgentCall says the agent’s intelligence runs locally on the user’s machine, including reasoning, memory, and decisions, while the service handles meeting plumbing and stores nothing long-term. It also says it never trains on user data. The page lists integration with Claude Code, OpenAI Codex, Cursor, Gemini CLI, Windsurf, GitHub Copilot, JetBrains Junie, OpenClaw, and the platform states that the agent frameworks developers build with and the languages they already know are supported.