Run every AI coding agent from one terminal with agent-manager, a terminal UI that manages Claude Code, OpenCode, Codex, and Grok Build sessions inside tmux. Each agent runs in its own tmux session, so work continues after you quit the manager, and you see live status, a foldable project tree, live pane preview, and resource gauges in a single list. Press space to send a prompt directly to a session or spawn a new agent in a selected group; use v to revive a dead session, Ctrl+Q to detach, and Ctrl+r to view a full‑file diff with syntax highlighting and line comments that go back to the agent.

Install via Homebrew (`brew install yoanwai/tap/agent-manager`), Go (`go install github.com/YoanWai/agent-manager@latest`), or prebuilt binaries; Windows users run inside WSL2. The manager checks GitHub Releases daily and shows an upgrade badge when a newer version is available. Key bindings include n for new session, g for new group, Enter to attach/fold, K/J to reorder, m to move, r to rename, a/u to archive/restore, d to delete, t to toggle archived view, e to hide empty groups, ? for help, and q to quit while sessions keep running.

Self‑naming sessions get placeholders like claude‑a1b2 and can be renamed via the agent‑manager rename command. Sessions can declare their working repo with `agent-manager review-repo <path>` and set a diff base with `agent-manager review-base <ref>`. An embedded MCP server exposes rename, review_repo, and review_base as native tools for MCP‑capable agents, while the CLI subcommands work everywhere.