ZCode

ZCode is an Agentic development environment from Z.AI that turns natural-language objectives into coordinated coding work. Rather than ACTing as a simple autocomplete layer, it is designed to operate insIDE a real project: interpreting requirements, proposing a plan, editing files, running commands, checking changes, and iterating until a task is complete. Its desktop workspace is built around the idea that a developer can act more like a project manager while an AI agent handles much of the execution.

What it does

ZCode connects a local project folder to an AI Agent that can read code, modify files, execute terminal commands, and present changes for review. The product combines a task-oriented interface with a project knowledge base, Git views, and configurable permissions. It is intended for work that spans multiple files or steps, such as adding a feature, fixing a bug, refactoring a module, or generating tests. The current 3.0 release emphASIzes a self-developed agent core and close integration with Z.ai’s GLM model family, including long-context options.

Key capabilities

• Goal-driven task execution: describe an outcome, then let the agent break it into steps and cARRy out code changes with review points.
• Grouped task workspace: manage several agent jobs at once, move tasks between stages, and pause or resume work in batches.
• Project Knowledge Base: scan an unfamiliar repository to generate structured notes about directories, modules, dependencies, and progress.
• Visual Git support: inspect branch relationships, switch branches, review diffs, and generate commit messages from inside the workspace.
• Extensible tooling: connect MCP services for Web Search, page reading, repository access, visual analysis, or internal tools.
• REMOte oveRSIght: monitor longer-running sessions from a mobile device and APProve sensitive actions when away from the desktop.
• Reusable Skills: capture repeatable Workflows so teams can apply consistent steps to common development tasks.

Who it is for

ZCode suits developers who want to delegate larger coding objectives and review the results, especially those working on multi-file changes, legacy code, or parallel maintenance tasks. It may also appeal to teams that prefer a Chinese-language development stack or want to keep model choices within the GLM ecosystem. Developers who mainly want instant inline completion, a mature debugger, or flexible switching among many model providers may find a traditional AI-native IDE or terminal agent more suitable.

How to get started

1. Download the desktop application for macOS or Windows.
2. Sign in and choose a local project directory as the workspace.
3. Let the project scanner build a knowledge base, then review the generated structure.
4. Create a task, describe the desired result, and choose a planning or direct-edit mode.
5. Review proposed diffs, approve sensitive commands, and comMIT changes through the built-in Git tools.

Pricing and access caveat

The desktop app can be downloaded without charge, while heavier use is tied to GLM Coding Plan subscriptions and token allowances. free models, trial credits, daily quotas, regional pricing, and plan limits may vary and can change over time, so confirm current terms on the provider’s site before relying on a specific allowance.