Fork of OpenAI's Codex CLI with a few additions:
- Subagents: Main session spawns specialized agents for exploration, execution, or research. Each streams in its own pane so you can watch tool calls and diffs in real-time.
- Persistent memory: Agent extracts knowledge from file reads, commands, and failures. Stores in SQLite per-project. Lessons and decisions never decay; other memories fade if unused.
- Live settings: Change config while the agent is mid-response. No need to restart the session.
- Codebase indexing: Integrates with sgrep for semantic search. Index builds in background, agent uses it for ranked code lookups.
Storage lives in .kaioken/memory/. Install via npm i -g @jayasuryajsk/codex-kaioken.
Subagents and memory add power, but also surface failure modes fast. In our experience, orchestration matters more than adding intelligence.