Kimi Code released versions 0.34.0 through 0.36.1 in August, focusing on improved agent orchestration, better task visibility, and enhanced workflow management.
Agents and Automation
- Subagent model pool: Configure a default model or named pool that spawned subagents select from, giving you control over agent behavior.
- WaitFor tool: Pause agents for background tasks to complete within the same turn instead of ending and restarting conversations.
- Live task progress: The /tasks panel now displays real-time work progress of background subagents with a card grid layout and status filtering.
- Tower mode: Experimental multi-agent orchestration capability, enabled via environment variable.
Interface and Workflow
- Plan and goal toggles: Access /plan and /goal commands directly in the input area; a new plan viewer panel lets you click entries to review full plans and results.
- Session management: Mark sessions as done to declutter your active list, and filter or batch-manage sessions across workspaces by status and update time.
- Multi-skill activation: Type / after whitespace to insert multiple skill tokens in one prompt.
- Background task control: Move running Bash commands or subagents to the background directly from the running card.
Platform and Features
- Persistent failure card: Resume interrupted conversations with one click when a model request fails.
- Remote Control: Access a local web session from anywhere by setting an environment variable and running a command.
Additional features and improvements were also released this month but are not listed here.