🤖 AI Summary
A new toolkit packages tmux-based session management for running and restoring many Codex CLI instances as a single “farm.” Install with the provided setup.sh (auto-installs tmux and multitail via common Linux package managers, populates $HOME/bin, and wires your shell), then use simple commands to add projects (codex-add), snapshot windows to a manifest (codex-save -> ~/.config/codexfarm/manifest.tsv), restore or attach after reboot/SSH (codex-restore -a, codex-resume), and monitor all panes/logs in real time (codex-watch). Each Codex pane logs to timestamped files under ${XDG_STATE_HOME:-$HOME/.local/state}/codexfarm/logs/, and a “board” session lets you quickly scan linked windows. The system supports non-attaching batch startup, auto-restore on login, and mobile-friendly modes (codex-watch --simple, mosh-friendly tips, tmux aggressive-resize).
For AI/ML engineers this is practical: it makes running many long-lived CLI model instances reproducible, automatable, and observable across reboots and remote sessions. Key technical details include a TSV manifest-based snapshot/restore flow (with -f to force recreate), environment-variable overrides (CODEX_CMD/ARGS/SESSION/NAME), multitail/tail modes for consolidated log viewing, and caveats like tmux’s inability to mirror the same live pane and pipe-pane only logging new output. The project is MIT-licensed and designed for typical Unix/bash workflows, easing multi-project experimentation, CI-style starts, and persistent remote development of Codex-powered tooling.
Loading comments...
login to comment
loading comments...
no comments yet