Rysh is a terminal multiplexer (like tmux or Zellij) written in Go where each pane runs an LLM agent that can answer prompts and execute tools. You get the familiar split-pane/tab interface, but the real difference is session persistence—agents resume mid-conversation after you stop and restart a session—plus coordination primitives like a shared board where agents post updates and read each other's work. It includes secret redaction (SecretNAT) that tokenizes credentials before sending requests to LLMs, keeping plaintext secrets local. The source is Apache-2.0; there's also a separate prebuilt "ry" distribution (closed binary). Installation is straightforward via `go install` or shell script, though Windows gets PTY support only through WSL2. The README includes five tutorial videos and documents the CLI commands for managing panes, agents, fleets (named agent groups), and inter-agent communication. One immediate question: when you run multiple agents in parallel and one fails partway through, does the board+fleet mechanism have built-in rollback or does orchestration still fall to the user?