Kimi Code released versions 0.21.0 through 0.31.0 in July 2026, adding major improvements to agent capabilities, file handling, reasoning controls, and the web interface.
Agents and Automation
- Expanded coder subagents: Subagents now support background tasks, todo lists, plan mode, skill invocation, and nested agents, matching the main agent's full capabilities.
- Custom agents as Markdown: Define custom agents as Markdown files with frontmatter and use them as main or sub-agents; plugins can also contribute custom agents.
- Secondary model configuration: New /secondary_model command lets you configure the model used by subagents.
- Print-mode background steering: Agent stays alive during background tasks and feeds completions back as new turns, with optional steering control via configuration.
Interface and Workflow
- Granular thinking effort levels: Select off, low, medium, or high reasoning effort directly from your editor instead of a simple toggle.
- File uploads and attachments: Accept any file type in chat; files the model cannot process inline upload to the server for on-demand access, appearing as chips in message bubbles.
- Session management improvements: Browse and restore archived sessions from Settings; search sessions by title, workspace, and prompt via Cmd/Ctrl+K palette with highlighted matches.
- Inline tool grouping and previews: Consecutive tool calls collapse into stacks with per-tool renderers, diff line counts, and inline previews for images, video, and audio.
Platform and Configuration
- Web server redesign: kimi web command runs in foreground, allows multiple instances on different ports, and supports pre-configured model credentials for ACP sessions.
- Session diagnostics and engine control: Export troubleshooting logs via /export command; switch between v1 and v2 engines at runtime in dev mode.
Several other features were also released this month but are not listed here.