dsh-zhipu-toolkit

DeepSeek Harness does not ship with a Zhipu GLM model-provider route out of the box. dsh-zhipu-toolkit fills that gap by registering the full GLM chat family (GLM-4.5 through GLM-5.3-Flash) as first-class models, so you can call Zhipu capabilities directly inside the same workspace without extra proxies or glue code.

The plugin exposes two billing channels — the Coding Plan subscription quota and the ordinary per-token API — with identical model lists and parameter semantics; the choice is purely which billing path your account uses. Core capabilities include a self-maintained GLM catalog with verified context windows and output caps, native image-input multimodality for GLM-5.3-Flash, fully local usage and quota statistics (requests, tokens, credits converted at official rates; nothing ever leaves the machine), live model discovery via /models polling at startup and after configuration changes, intelligent reasoning-tier mapping for GLM-5.3-series models (the API rejects thinking: disabled, so the plugin maps tiers automatically), and a hot-reloadable Web settings card.

Engineers who do daily code generation, reasoning debugging, or agent orchestration in DeepSeek Harness and already hold a Zhipu BigModel account will find this a drop-in addition. API keys can be stored in the local credentials store, configuration changes take effect immediately without a restart, and the entire pipeline works out of the box.