// the find
herdrdev/herdr
agent multiplexer that lives in your terminal.
herdr is a terminal multiplexer built specifically for running multiple AI coding agents in parallel — think tmux, but with first-class awareness of agent state (blocked, working, done). It's a single Rust binary that lets you split panes, detach and reattach sessions over SSH, and exposes a socket API so agents themselves can spawn panes and coordinate with each other. The target is anyone running Claude Code, Codex, or similar tools at scale and tired of juggling terminal tabs.
The socket API for agent-to-agent coordination is the genuinely interesting part — agents can spawn child panes, read output, and wait on each other without any external orchestration layer. Session persistence that survives restarts and works over SSH is done right: it's not a wrapper around tmux, it's its own session model. Shipping as a single Rust binary with no Electron is the correct call; it runs in whatever terminal you already have without a separate runtime. The project has real documentation structure with versioned docs, a plugin marketplace, and multi-language READs — this isn't a weekend hack.
Windows support is explicitly labeled 'beta', which is a real gap if your team isn't all on Mac/Linux. The plugin system is young — the marketplace exists but the extension API is TypeScript-based (see `.pi/extensions/`), which is an odd choice for a Rust project and adds a JS runtime dependency that the 'no Electron' pitch implicitly promises you won't need. At 22k stars with no 1.0 release yet, the API surface (especially the socket API) is likely to break between minor versions; the versioned docs directory going back to 0.5.11 suggests it already has. The agent skill system (`.agents/skills/`) is interesting but underdocumented — it's not obvious how to write or distribute skills beyond the two examples in the repo.