ToolStack

Z.ai Launches ZCode, a Coding Agent for GLM-5.2

Z.ai Launches ZCode for GLM-5.2 illustrated as an AI coding agent working on a real software project.

Z.ai released ZCode. It’s a desktop app built specifically to put its GLM-5.2 model to work on real coding projects, not just answer questions about code. ZCode is a full agentic development environment. It plans, writes, tests, and reviews code across long, complex tasks without losing track of what it’s doing partway through.

What’s New

  • ZCode runs everything through a single “ZCode Agent.” It keeps goals, open files, terminal output, browser context, and Git state together in one task. That means a long coding job doesn’t lose continuity halfway through
  • You give it instructions in plain language. It handles coding, debugging, testing, previewing the project, and reviewing changes on its own
  • It leans on GLM-5.2’s 1-million-token context window. That lets it keep reading files, terminal results, and project state continuously instead of forgetting earlier context as a task runs longer
  • You’re not stuck at your desk to use it. ZCode supports remote control from mobile, plus bot integration through Feishu, WeChat, and Telegram. You can check progress or add instructions while a task keeps running in the background
  • Sensitive actions go through a confirmation step first. File changes and high-permission commands don’t run silently
  • Z.ai is running a limited-time deal through July 31, 2026. GLM Coding Plan subscribers get roughly 1.5x their usual quota using ZCode. New users get a 5-day free trial with 5 million tokens a day to try it out

Why It Matters

Z.ai is doing something a lot of model makers eventually do. Once a model gets good enough at a specific job, they build a dedicated tool around it instead of leaving developers to wire it in themselves. GLM-5.2 already tests well on real coding benchmarks. ZCode is Z.ai’s bet that a purpose-built app, one that manages context, permissions, and long-running tasks on its own, gets more real use out of the model than the API alone.

The remote and bot control angle is worth a second look too. Letting an agent keep working while you check in from your phone or a chat bot sounds like a small feature on paper. It actually points at something bigger. Coding agents are starting to be built less like a tool you sit and use, and more like something you delegate to and check back on later.

Should You Care?

If you write code and you’re curious about long-context coding agents, yes. This is worth trying while the free trial and quota bonus are live. ZCode is free to download. The 5-day trial gives you 5 million tokens a day, enough room to test it on a real project instead of a toy example.

If you’re not doing hands-on development work, this is more of a signal than something to act on. It’s another example of AI companies moving past chat interfaces and into tools built to finish a job end to end. The same shift is showing up across coding, computer use, and agent-based tools right now.

Source: Z.ai: ZCode Documentation

Scroll to Top