Introducing Codex Kaioken – the Codex CLI fork with subagents, plan mode UX, indexing and manual checkpoints and restoring.
Posted by No-Point1424@reddit | LocalLLaMA | View on Reddit | 1 comments
I’ve been missing richer UX in the default Codex CLI, so I forked it into Codex Kaioken. It keeps all the upstream features but adds:
- Real-time subagent panes that stream tool calls, diffs, and timers as they happen
- Plan-first mode (toggle with /plan or Shift+Tab) with a cyan composer and feedback loops before execution.
- A /settings palette to adjust plan granularity, footer widgets, and subagent concurrency without editing config files.
- Checkpoint snapshots (/checkpoint save|restore) plus instant /undo
- An upgraded welcome dashboard showing branch/head, sandbox mode, rate limits, indexing status, and writable roots.
Source + docs: https://github.com/jayasuryajsk/codex-kaioken
It can be installed with
npm install -g /codex-kaioken
I’d love feedback especially on multi-agent UX ideas and the plan mode flow , any bugs or ux issues.
Restoring checkpoints is buggy and fixing it now.
bfroemel@reddit
very nice - thanks for your effort + will check this out!
Two comments right away:
- there is imo no need to merge or even keep every upstream feature; only focus on those that would really help you/the community you want your fork to be relevant.
- currently upstream supports that you can use your own models; incl. local ones. It would be great (well, at least for members in this sub here :) if you never remove that feature from your fork and/or even make it more easy to use local models.
Best of success!