ZCode - The official harness for GLM-5.2
ZCode is the official agentic development environment for GLM-5.2, built for long-running coding tasks with stable context, file edits, terminal and browser state, Git review, mobile Remote, bot control, BYOK, and macOS, Windows, and Linux apps.
Replies
Flowtica Scribe
Hi everyone!
Have been using ZCode since its first version late last year. With the recent 3.0 upgrade, it's taken a massive leap forward. Here are a few reasons it’s worth a try:
It’s the official harness for GLM-5.2. Yes, try the tuning that understands the true upper bound of GLM-5.2 better than anything else!
It’s built for long-horizon execution. The in-house ZCode Agent keeps your files, terminal output, browser context, and Git state stitched together in the same task loop, so it doesn't lose the thread halfway through a complex build.
It lets you step away from the desk. You can steer the agent, check progress, and kick off tasks from your phone, or directly through messaging bots while the desktop keeps running.
And it’s moving fast!
the minimalist UI actually lets the models breathe instead of competing with them. nice to see a platform that trusts the work to speak for itself.
Curious how the rumination model handles long context sessions compared to the base version?
Congrats on the launch! 🚀
Long-running coding agents are where things usually get difficult, especially when the agent has to keep file state, terminal output, browser context, and Git changes aligned across one task.
Curious how ZCode handles situations where the agent gets stuck or starts making risky edits. Does it surface checkpoints or risk summaries before continuing?
The minimal UI really lets the models speak for themselves, no clutter or upsells getting in the way of actually trying them out.
Reasoning mode actually explains its thinking out loud, which helped me catch where it went off track. Nice to see solid open weights like this finally have a clean free playground to test in.
Tried the reasoning model on a tricky coding question and it walked through the logic clearly instead of just spitting out an answer. The bare-bones UI is kind of refreshing honestly.
The reasoning model handled a tricky coding question without any hand-holding, which caught me off guard. Clean interface, no fuss, just straight to the response.
Congrats on the launch, MIT licensing is a nice touch. Quick question though, are there any rate limits or token caps per day for free users right now, or is the unlimited angle going to hold once traffic picks up?