Developer tombaldwin has released agentlink, an open-source tool that allows Claude Code and Codex agents to communicate across different user accounts and machines. While Claude Code’s native messaging features are powerful, they are constrained to a single account or require Anthropic’s remote control infrastructure. Agentlink bypasses these limitations by establishing a direct, encrypted mesh network between local machines, enabling true multi-agent orchestration without central servers.

Bypassing Account Silos with Direct P2P

The core innovation here is the ability to address agents using a name@host format, such as reviewer@laptop.local, regardless of which Claude account owns the session. This allows for complex workflows where a code reviewer on a desktop machine can wake up an idle session on a laptop, even if they belong to different Anthropic accounts. The system uses encrypted UDP beacons for zero-config discovery on local networks and supports Tailscale for seamless connectivity across different physical locations.

Robust Delivery and Security Mechanisms

Agentlink employs a store-and-forward mechanism to ensure message reliability. If a target machine is offline, messages are stored on disk and retried, with copies held by up to three other online nodes in the network. Security is handled via per-machine keys and TLS encryption, ensuring that messages are signed and cannot be altered by relaying nodes. The tool also mitigates prompt injection risks by instructing agents to treat incoming messages as requests from colleagues rather than direct user instructions, requiring confirmation for destructive actions.

Key Takeaways

  • Supports Claude Code 2.1.318+ and Codex 0.167+, with extensibility for other MCP agents via polling.
  • Requires Apple Silicon Macs or Linux; Intel Macs are currently unsupported.
  • Features automatic discovery on LAN and integration with Tailscale for remote access.
  • Messages expire after 7 days, with configurable storage limits to prevent runaway agent loops.

The Bottom Line

Agentlink is a crucial piece of missing infrastructure for developers running multi-machine AI workflows. It solves the fragmentation problem of account-based silos, enabling a decentralized, resilient agent mesh that feels like a natural extension of the local development environment.