ZCode is Z.ai’s desktop Agentic Development Environment built around GLM-5.2 and long-running coding-agent workflows. Official docs describe it as a way to bring GLM-5.2’s long-context, long-horizon and agentic coding capabilities into a stable workspace for planning, coding, reviewing and iterating across complex development tasks. Unlike a simple autocomplete plugin, ZCode is positioned as an end-to-end agent workspace: users pick a project directory, connect Z.ai, BigModel or an API key, describe a goal, reference files, run commands, manage Git context and keep the agent working through a desktop interface.
The public source base is sufficient for a source-clean tool page. The official site and docs were re-accessed during the July 5 write check, with the homepage available at zcode.z.ai and English docs available at /en/docs/welcome. The install page lists v3.2.5 as the latest desktop release, with macOS Apple Silicon/Intel and Windows x64 support plus a Linux beta path. The changelog shows multiple July 3 releases, including fixes for long agent tasks, MCP workspace trust defaults, slash-menu skills, authentication, conversation restore and Linux update behavior. Core docs cover the ZCode Agent, Goal Mode, Remote Control, MCP Servers, Usage Stats and Safety Confirmation.
The CMS page should present ZCode as a Cursor, Claude Code and Codex-adjacent IDE alternative for developers evaluating GLM-powered coding agents, especially teams interested in long-horizon tasks and cross-device supervision. Key differentiators to describe are Goal Mode, which tracks a verifiable objective across iterations; Remote Control, which lets a phone send instructions to the active desktop workspace; MCP server management; and permission modes such as Confirm Before Changes, Auto Edit, Plan and Full Access. Caveats: exact pricing needs a current official pricing check, Linux support is marked beta, and GLM-5.2 capability claims should be attributed to Z.ai docs rather than independent testing.
