Herdr is an open-source, AI-native terminal multiplexer engineered in Rust, designed specifically to coordinate and manage workflows across multiple AI coding agents running in parallel. While traditional multiplexers like tmux and Zellij treat terminal panes as agnostic text streams, Herdr introduces agent awareness by continuously parsing process state, detecting whether an agent is active, idle, waiting for user confirmation, or completed. This enables developers to run tools like Claude Code, OpenAI Codex, Aider, OpenCode, and Pi in isolated panes while monitoring their operational state from a unified interface.
Beyond session multiplexing, Herdr functions as an active control plane for multi-agent collaboration. The platform exposes a local Unix domain socket API and a companion CLI that agents can query to inspect sibling panes, spawn subordinate helper agents, pass context across workspaces, and wait on asynchronous task completions. Through reusable agent skill files and Model Context Protocol (MCP) tool bindings, running coding assistants can autonomously delegate subtasks—such as offloading test execution or code review to a separate pane—without leaving the terminal environment.
Herdr operates on a client-server daemon architecture where agent sessions and terminal buffers persist independently of the active terminal window or SSH connection. Developers can safely detach from long-running agent workflows, close their terminal emulator, and reattach seamlessly across local sessions or remote SSH environments without interrupting active tasks. The interface offers both mouse-driven controls (pane resizing, drag-and-drop tab management, and visual status sidebars) and configurable keyboard shortcuts, bridging traditional CLI workflows with modern GUI ergonomics.
Herdr is distributed as free, open-source software under the Apache-2.0 license, with prebuilt binaries and packaging available for Linux, macOS, and Windows. Because Herdr runs locally on developer workstations or self-hosted development servers, it imposes no external telemetry, mandatory cloud dependencies, or additional platform fees. Infrastructure and model inference expenses are determined entirely by the individual AI tools and provider API keys configured by the user.
