ZCode’s own repository labels it an AI coding workbench whose components cover the desktop, the browser and terminal agents. Zhipu has now put that code on GitHub, pairing the release with an apology over a data-handling dispute and word that a remediation process has finished.
Publishing the source answers one of the complaints raised after reports the tool uploaded local data, an episode that drew scrutiny from Chinese regulators.
Retention is the second half of the announcement. Zhipu’s model-as-a-service platform will soon let users apply to keep their data out of retention, the company said, and ordinary model calls will not be stored persistently – inputs and outputs are to serve only the request in flight.
The carve-outs deserve a close read, because this promise is far from blanket. Traffic on the Batch API and the File API can still be kept for a defined period, and so can material the company holds for legal, security or abuse-prevention reasons. Promises of that shape are standard across enterprise AI, and they leave plenty of room for retention in practice.
Zhipu, which trades overseas as Z.ai, ranks among China’s better-funded model developers, carrying a Hong Kong listing and heavy compute spending behind it. Capability is not the hard sell abroad; trust is. A developer weighing a Chinese coding agent against a US alternative wants to know where the source code ends up.