👉 Version compare: v0.7.15...v0.8.0

v0.8.0

kimi

Added

• Added kimi module —— because I bought a 7-day trial for 7 yuan.

  Since I'm using it anyway, might as well make it smooth. I built a scaffold to deploy it quickly across different servers and workstations.
  No manual dependency juggling or environment setup — one command and you're good to go.

  Trial expiring? Uninstall is one command too, clean and simple.

  Examples:

  # Launch Kimi Code, auto-installs via uv if not present
  x kimi
  
  # Upgrade kimi-cli to the latest version
  x kimi --upgrade

claude

Added

• Added attribution|attr subcommand —— honestly, I built this because it was bugging me.

  Here's the thing: I recently started using Claude Code, but often with DeepSeek models.
  Yet every time, there's Claude's icon sitting right there in the co-author list.
  I just don't like it.

  Don't get me wrong, I believe the Claude Code team meant well —— transparency about AI-assisted code is good practice.
  And they did provide a way to turn it off (though it's a bit buried).

  So I made a one-click tool with three options:

  • Remove it entirely: x claude attr rm
  • Keep it generic as Co-Authored-By: AI
  • Or customize it to whatever you want

  Examples:

  # Straight up: I don't want attribution on every commit
  x claude attr rm
  
  # Or set your own
  x claude attr use --msg "Co-Authored-By: deepseek <noreply@x-cmd.com>"

• Added mt subcommand —— Moore Threads dropped a developer bundle too: 30-day free trial of their lite tier, powered by GLM 4.7.

  Haven't snagged my own invite yet, but the scaffold's ready.
  When you get in, one command hooks Claude Code up to domestic GPUs.

  More players, more options. x-cmd keeps tabs on these deals (and keeps grabbing them).

  Example:

  # Claude Code + Moore Threads, one shot
  x claude mt

• Added create subcommand —— we're collecting best practices for Claude Code workflows.

  Truth is, we're vibe coding beginners too, still figuring out what actually works.
  This experimental release is rough around the edges, give us time to iterate.

  If you're willing to be a guinea pig, try it out. But manage your expectations.

doubao

Added

• Added doubao module —— truth is, I haven't started using it myself yet.

  But users asked for it, so I built the scaffold.
  When you're ready, one command gets you on Volcano Engine's Doubao model.

  Example:

  # Interactive setup
  x doubao init
  
  # Call Doubao directly
  @doubao "Give me an example of recursion in Python"

gg

Added

• Added gg module —— spun it out from x gemini gg because I actually use this thing.

  Free Google search + Gemini, pretty good quality. But here's the catch: Google's references are all wrapped in intermediate pages, used to fetch them one by one, slow as hell.
  Spent an evening adding concurrency. Now it's under 10 seconds.

  Gemini's free tier is generous and the quality's solid, plus you're waiting for AI anyway. Works for me.

  Planning to package this as a skill for various agents. If you've got cooler use cases, share them.

  Example:

  # Let Gemini Google for you
  x gg "What is x-cmd?"

gram

Added

• Added network kill switch to x gram stop 2 —— honestly, I'm kind of worried.

  Unrestricted agents might already be lurking and attacking. clawdbot and moltbook are heating up right now —— we just launched a response site this Sunday.
  We threw up a manifesto site in a hurry: https://bot-killer.x-cmd.com/

  Last version we shipped "kill by process name" and "memory file checkpointing".
  This version, x gram gets real teeth: one command severs all HTTP/HTTPS connections, killing processes by their network links.
  Stops malicious agents from "calling for backup" or silently exfiltrating your local data.

  Also added stop 3/4/5 strategies:

  • stop 3 —— Based on stop 2, archive and remove all folders in $HOME containing soul.md and memory.md
  • stop 4 —— Based on stop 3, additionally kill all processes using these memory folders. Dual protection with "kill by name" and "kill by file"
  • stop 5 —— Based on stop 4, expand search scope to entire / directory (only removes *.md files in system directories)

  Next step? I'm working on using AI to help users identify these threats.

zhipu

Added

• Added support for glm-4.7 and glm-4.6 —— same story, haven't had time to really use them myself.

  But users wanted it, so it's there. Pick your version with --model.

  Example:

  @glm --model glm-4.7 "As a marketing expert, create a catchy slogan for my product"

chat

Fixed

• Fixed JSON Unicode handling in x-cmd agent —— hit this while testing the Zhipu API.

  Certain Unicode characters were causing request JSON to choke. Patched.

zuz

Fixed

• Fixed zuz stability issue.

  We actually knew about this —— someone aliased pwd, and zuz was still using the old command.
  zuz code has been rock solid, untouched for years. When @polymerase60053 filed the issue, we rewrote this part with a better approach.

  Thanks for the nudge, @polymerase60053! #370 (comment)