One thing I intentionally left open: tmux as the orchestration layer
means Mimir isn't Claude Code-only. Since agents are just processes
in tmux panes, you could run Codex in one pane, Grok in another,
Claude Code in another — all sharing the same memory layer via Mimir.
Multi-model swarms aren't on the roadmap yet, but the architecture
already supports it.
Multi-model swarms aren't on the roadmap yet, but the architecture already supports it.