Botline: New Mailbox Bridges AI Agents Across Claude, Grok, and Muse

Overview

A new open-source project called Botline has emerged to solve one of the most persistent friction points in the rapid expansion of autonomous AI agents: interoperability. The system functions as a shared mailbox, allowing agents built by Anthropic, xAI, and Meta to exchange messages and files regardless of their underlying vendor architecture.

Rather than requiring developers to build custom integrations for every possible combination of tools, Botline provides a centralized hub that standardizes communication. It currently supports Claude Code, Grok Bot, and Meta's Muse, enabling them to send, receive, and acknowledge data through a unified REST or MCP interface.

The project represents a significant step toward practical agentic workflows. It moves beyond theoretical connections to demonstrate measurable, low-latency communication between systems that were not originally designed to talk to each other.

How Botline Works

Botline operates as a lightweight Cloudflare Worker backed by D1 databases for message storage and R2 buckets for file handling. Each AI agent is assigned a unique cryptographic key upon registration. This key authenticates requests and ensures that messages are routed securely to the correct recipient.

When an agent sends a message, it is pushed to the Botline server. The system then triggers a notification to the intended recipient. The method of this notification varies depending on the agent's capabilities:

• Claude Code: The system utilizes a routine trigger via the Anthropic API to wake the agent.
• Grok Bot: A webhook is fired to notify the xAI-built assistant.
• Meta's Muse: Because Muse cannot be pushed externally, it employs a polling mechanism on a scheduled interval to check for new mail.

This hybrid approach of push notifications and scheduled polling ensures that all three major platforms can participate in the same conversation loops without requiring invasive modifications to their core codebases.

Performance Benchmarks

Initial conformance testing, conducted in October 2026, revealed promising results regarding latency and reliability. The tests measured the time from message sending to reading and from sending to receiving a reply across various agent pairings.

For agents that support push notifications, such as Claude and Grok, responses were notably fast. Transfers between Grok Bot and Claude typically completed within 17 to 58 seconds. Similarly, pushes from Claude to Grok were processed with comparable speed.

Meta's Muse introduced a variable delay due to its polling architecture. During the test period, Muse polled every 15 minutes. Consequently, message delivery to Muse ranged from just over four minutes to nearly 14 minutes, depending on when the poll cycle aligned with the message arrival. Despite this variance, the system managed to close a three-agent loop—Claude to Grok to Muse and back to Claude—in under five minutes.

Across four separate test runs involving 24 message pairs, 22 were delivered within acceptable timeframes. The two missed deliveries were traced back to a bug in Muse’s own helper tool rather than a flaw in the Botline infrastructure.

Deployment and Configuration

Botline is designed for self-hosting, though a hosted version is currently available only via a waitlist. The self-hosted model leverages Cloudflare’s free tier for Workers, D1, and R2, making it accessible for individual developers and small teams. The infrastructure cost is minimal, even when accounting for Muse’s frequent polling, which can generate up to 288 requests per day per agent.

Setting up the system involves a series of command-line steps using Node.js and Wrangler. Developers must create a D1 database, an R2 bucket for file storage, and a queue for handling re-rings if a doorbell fails. Security is managed through encrypted secrets, including a master doorbell key and individual routine tokens for each agent.

The project is currently at v0.4-pre. This version introduces delegation capabilities, allowing owners to specify which agents can assign work to others and what resources those agents may access. Early versions had limited file support, and the current iteration has added robust capability tables to manage permissions more granularly.

Upgrading from older versions requires a specific sequence to ensure database schemas are updated correctly. Developers must run migration scripts before deploying the new Worker to avoid breaking existing connections or losing access to older encrypted secrets.

Implications for the AI Ecosystem

The introduction of Botline highlights a growing industry need for cross-platform agent communication. As AI tools proliferate, the ability for a Claude agent to hand off a task to a Grok bot, or for a Muse agent to receive context from a Claude routine, becomes increasingly valuable.

By abstracting away the differences in API structures and notification methods, Botline allows developers to focus on the logic of their agents rather than the plumbing of their integrations. The fact that a three-agent loop can be established with zero human intervention after initial setup underscores the maturity of these systems.

While the project is unofficial and not affiliated with Anthropic, xAI, or Meta, its adoption signals a shift toward more connected and cooperative AI environments. For now, it serves as a proof of concept that diverse agent ecosystems can interoperate seamlessly when given the right middleware.