Kimi CLI v1.40.0 Changelog

This release is all about making the Kimi command-line tool more reliable and easier to use. It matters most to developers and power users who use this tool for heavy, automated tasks, because it fixes several connection and login glitches that used to interrupt workflows, while raising the limits on how much the AI can do in a single turn.

What's new

• Session tracking and telemetry: The app can now collect and send usage statistics (telemetry) to a dedicated server, helping developers understand how the tool is being used so they can improve it.
• AFK (Away From Keyboard) mode: There's a new mode that handles things independently when you're not around, which works alongside the existing "YOLO" auto-approve mode without interfering with it.

Improvements

• Higher step limit: The default maximum number of steps the AI can take in a single turn was doubled from 500 to 1000. This means the AI can tackle much bigger, more complex tasks before it has to stop and ask what to do next.
• Better testing: The developers added "end-to-end" accuracy tests. This is an automated testing setup to ensure the AI's responses stay high-quality and don't break in future updates.

Fixes

• Login recovery (OAuth): Fixed an issue where a temporary network hiccup could break your login flow. It also now properly saves your refreshed login token if the connection drops and reconnects.
• Usage quota display: The command that shows how much of your usage limit is left (/usage) was showing the wrong numbers. This is now fixed.
• Task counter in status bar: The status bar at the bottom of your screen now correctly shows how many active background tasks the AI agent is running.
• YOLO mode freezing: In "YOLO" mode (where the AI skips asking for your permission to run things), the AI would sometimes freeze if it actually needed to ask you a question. That block is now cleared.
• Context compaction reminder: When the AI's memory gets too full, it "compacts" (summarizes) the conversation to save space. It used to forget its safety reminders after doing this, but now it re-injects them so it stays safe.
• Approval request lifecycle: Fixed a bug where the AI would ask for your permission to do something, but the request would linger or get mixed up between different turns of the conversation. Now, permission requests are strictly tied to the current turn.
• Echoing slash inputs: When you typed a slash command (like /usage) in a workflow, it wasn't showing up properly on your screen. Now it echoes back correctly.
• Web media previews: Fixed a glitch in the web interface where previews of files or media (like images) generated by tools would randomly disappear.
• Web session titles: Fixed an issue where the web interface would overwrite the custom names you gave to your chat sessions.

Breaking changes

No breaking changes in this release.