Agentring Project Launches Open-Source Tool for AI Agents to Place and Receive VoIP Calls

An open-source project named Agentring has been introduced, allowing AI agents to initiate VoIP phone calls to users and receive inbound calls. According to the available documentation, the system is designed to notify users when an agent encounters an issue, requires a decision, or completes a task, making the phone ring like a traditional call even when the device is locked.

Core Architecture and Features

Agentring is self-hosted and operates without traditional phone numbers or third-party telephony services like Twilio. Calls utilize VoIP infrastructure running from a user's server to their device. The system comprises several components:

• Relay (TypeScript): Handles pairing, messages, call states, push notifications, and LiveKit tokens, typically running on a Linux server behind Caddy.
• LiveKit: Manages call audio and is self-hosted within the same Docker Compose configuration.
• Daemon (Python): Contains the command-line interface, Model Context Protocol (MCP) server, and voice AI integration.
• Android App (Kotlin): Provides QR pairing, a native call interface, and message handling.

Operational Workflow

The software allows bidirectional communication between users and AI agents. Agents can send messages, await replies, or trigger phone calls using specific functions. During a call, a voice AI explains the agent's status, addresses questions, and returns a transcript to the agent. Conversely, users can initiate calls to their agents via the mobile application, and the voice AI can consult other agents during an active conversation.

Requirements and Considerations

Running Agentring requires a server or local computer with Docker, a free Firebase project to handle push notifications on locked devices, and an API key for a supported voice AI provider such as Gemini, OpenAI, xAI, or Phonic.

Documentation notes certain limitations and security considerations. The software is currently in pre-release and has been tested primarily on an Android 16 device (Pixel 6). Users should review security documentation regarding tool permissions and data handling, as call audio and metadata pass through configured servers and third-party AI providers depending on the selected settings.

Verification and Availability

Interested parties can review the source code, setup tutorials, and contribution guidelines via the official GitHub repository maintained under the mutonby/agentring namespace.