Herdr: Terminal Multiplexing for AI Coding Agents

Agent-Native Multiplexing

Herdr functions as a terminal multiplexer specifically designed for AI coding agents, providing visibility into agent states that standard tools like tmux lack. It tracks four distinct states for each agent: working, blocked (waiting for user input), idle, and done. The interface includes a sidebar that displays these states in real-time, eliminating the need to manually cycle through terminal panes to check progress. It supports approximately 20 agent CLIs out of the box, including Claude Code, Cursor Agent, and Gemini CLI, using either direct lifecycle hooks or terminal buffer heuristics to detect state changes.

Automation and Orchestration

Beyond monitoring, Herdr exposes a CLI and a local socket API that allows agents to coordinate with one another. Agents can programmatically split panes, launch other agents, send prompts, and wait for specific states or outputs. This enables complex workflows, such as having a primary agent trigger a secondary agent to perform a code review on a generated diff, wait for the review to complete, and then ingest the feedback automatically. The tool also supports remote mode, allowing users to run agents on a remote server or VPS while maintaining local key bindings and session persistence.

Persistence and Configuration

Herdr utilizes a client-server architecture to ensure sessions survive disconnects or server restarts. For supported agents, it can automatically resume previous conversations upon reconnection. Configuration is managed via a local file, allowing users to customize themes, key bindings, and notification settings. While the tool is currently pre-1.0 and relies on heuristics for agents lacking native hooks, it provides a stable mission control for developers running multiple AI agents in parallel.