# AgentCall

> Your AI agent, in every meeting.

AgentCall is a skill that turns AI agents into teammates who join video meetings on Google Meet, Zoom, and Microsoft Teams. The agent talks, listens, presents, screenshares, and collaborates — just like a real teammate. It is distributed as a skill (compatible with the open Agent Skills standard), not a client library — agents install the skill once and then participate in meetings on a user's instruction.

## What AgentCall Does

AgentCall lets AI agents:
- Talk: speak naturally in meetings with built-in voice generation, interruption handling, and barge-in prevention
- Listen: receive real-time transcription with speaker identification (currently English, more languages coming soon)
- See: take screenshots of shared screens, slides, and presentations
- Present: screenshare dashboards, slides, and demos dynamically during calls
- Chat: send and read meeting chat messages (URLs, code, links)
- Work: search code, edit files, run commands, and build — all while staying in the meeting

## How It Works

1. Install the skill — easiest path: ask your agent to "install join-meeting from github.com/pattern-ai-labs/agentcall". For Claude Code specifically: `/plugin marketplace add pattern-ai-labs/agentcall` then `/plugin install join-meeting@agentcall`. Skill is compatible with Claude Code, OpenAI Codex, Cursor, Gemini CLI, Windsurf, JetBrains Junie, OpenClaw, and any other framework that supports the open Agent Skills standard.
2. Sign in once — your agent will prompt for your AgentCall API key on first run. Get it from https://app.agentcall.dev/api-keys. Saved to `~/.agentcall/config.json` and reused across sessions.
3. Tell your agent to join — "Join this meeting: https://meet.google.com/abc" — the agent talks, listens, codes, builds, answers, and multitasks in the call.

## Modes

- audio: voice only — for standups, 1-on-1s, support calls
- webpage-audio: voice + custom HTML rendered as the bot's video feed
- webpage-av: voice + animated avatar — for team meetings, customer-facing calls
- webpage-av-screenshare: voice + avatar + screenshare — for presentations, training, reviews

## Voice Strategies

- direct: agent controls TTS directly — best for 1-on-1 conversations
- collaborative: GetSun voice intelligence handles real-time conversation timing (trigger words, barge-in prevention, interruptions) — best for group meetings

## Pricing

Billed per minute. No minimums. No subscriptions. 6 hours of meeting time included on signup.

Base rates (per hour):
- Audio mode: $0.35/hr
- Avatar mode: $0.45/hr
- Avatar + Screenshare mode: $0.50/hr

Add-on rates (per hour, on top of base):
- Transcription (STT, English): $0.12/hr
- Voice intelligence (GetSun, collaborative mode only): $1.00/hr
- Text-to-speech (Kokoro TTS, per generated audio): $8.00/hr

Plan tiers:
- Base: pay-as-you-go from $0.35/hr, 1 concurrent call, 1 hour max call duration
- Pro: discounted base rates, more concurrent calls, up to 4 hour max call duration
- Enterprise: lowest base rates, unlimited concurrent calls, up to 8 hour max call duration

Current rates: https://agentcall.dev/#pricing

## Supported Platforms

Meeting platforms: Google Meet, Zoom, Microsoft Teams (both teams.microsoft.com work/365 and teams.live.com consumer/free).

Agent frameworks: Claude Code, OpenAI Codex CLI, Cursor 2.5+, Gemini CLI, Windsurf, JetBrains Junie, OpenClaw, and any other framework that supports the open Agent Skills standard.

Languages: Python 3.10+, Node.js 18+ (bridge scripts included).

## Getting Started

1. Sign up: https://app.agentcall.dev/login (Google OAuth or email OTP, free)
2. Get API key: https://app.agentcall.dev/api-keys (or use the Quick Install onboarding page that opens for new users)
3. Install the skill: https://github.com/pattern-ai-labs/agentcall
4. Tell your agent to join a meeting

## Key Features

- VAD gap buffering: combines fragmented transcripts into complete utterances
- Barge-in prevention: waits for silence before speaking
- Auto-interruption detection: stops speaking when someone talks over the bot (webpage modes)
- Sentence tracking: knows which sentence was interrupted and when
- WebSocket reconnection: auto-reconnects on network blips
- 7 built-in UI templates (pattern, orb, ring, avatar, dashboard, blank, voice-agent)
- Screenshots: capture what's on screen at any time
- API key persistence: saved to `~/.agentcall/config.json`, asked once
- Stop Hook for Claude Code: prevents agent from exiting mid-call

## Links

- Website: https://agentcall.dev
- Dashboard: https://app.agentcall.dev
- GitHub: https://github.com/pattern-ai-labs/agentcall
- YouTube: https://www.youtube.com/@pattern-ai-labs
- X (Twitter): https://x.com/patternailabs
- Blog: https://patternai.substack.com/
- Contact: hi@agentcall.dev | +1 (415) 430-7518

## Technical Details

- Distribution: open Agent Skills standard (SKILL.md + bridge scripts in `scripts/python/` and `scripts/node/`). Not a client library — humans do not import a Python or Node SDK.
- API surface (consumed by the skill, not directly by humans): REST + WebSocket at api.agentcall.dev. Auth via API key (`ak_ac_` prefix) over HTTPS / WSS.
- Events (delivered to the agent in real time): transcript.final, transcript.partial, participant.joined, participant.left, tts.done, tts.interrupted, call.bot_joining, call.bot_ready, call.ended, etc.
- Commands (issued by the agent): tts.speak, send_chat, raise_hand, mic, screenshot, webpage.open, tasks.set, set_state, screenshare.start, screenshare.stop, leave.

## About

Built by Pattern AI Labs (https://pattern-ai.com).
License: MIT (skill package).