As developers push the limits of agentic coding, the bottleneck has shifted from model capability to human context management. Relaymote, a new unofficial tool released on November 16, 2026, addresses this by introducing a local 'Conductor' architecture that routes requests across multiple Claude Code sessions. The tool, developed by Ashwar Sadh, operates entirely on the local machine using the official Claude Desktop app, eliminating the need for a proprietary cloud service layer.

The Three-Tier Orchestration Model

Relaymote structures workflows into three distinct roles: Workers, Masters, and the Conductor. Workers are standard Claude Code sessions focused on single tasks. Masters coordinate specific projects, spawning workers on appropriate models and tracking their status. The Conductor sits at the top, holding a global project index and routing incoming requests to the correct Master. This hierarchy ensures that the Conductor’s context window remains small and efficient, as it strictly handles routing and indexing rather than executing code itself.

Routing Logic and Model Mapping

The system’s core innovation is its routing mechanism, which uses tools like relaymote_projects and relaymote_route_owner to direct queries. Unlike manual session management, where developers might paraphrase requests, Relaymote passes the user’s exact words to the owning session if it exists, preserving context fidelity. If no owner exists, the Conductor initiates a new Master in the correct folder. Furthermore, the tool allows for granular model mapping, assigning smaller, cheaper models to 'easy' tasks and reserving high-effort, large models for 'hard' tasks, optimizing both cost and performance.

Transparency and Audit Trails

Every action taken by a Master or the Conductor is logged locally in an audit trail, providing transparency into the agent’s decision-making process. The system includes recovery tools such as relaymote_resume, relaymote_unstick, and relaymote_heal to manage stuck sessions. Crucially, Relaymote does not autonomously decide when a task is complete; it requires evidence from the session and user verification before closing goals, maintaining human oversight in the loop.

Key Takeaways

  • Relaymote is an MIT-licensed, local-only tool that works with the official Claude Desktop app, currently supporting Windows and experimental macOS.
  • The architecture prevents context bloat by ensuring the Conductor session only routes requests and manages the project index, never executing code.
  • Model selection is dynamic, with Masters spawning Workers on specific models based on task difficulty, optimizing token usage for complex projects.
  • The tool is explicitly unofficial and not affiliated with Anthropic, relying on the existing Claude Code ecosystem.

The Bottom Line

Relaymote offers a pragmatic, local-first solution to the chaotic reality of managing multiple AI agents, proving that better orchestration often beats raw model power.