Introduction¶

DeepSeek Harness (DSH) passes the provider’s model list to the bundled pi-ai catalog for parsing. If Zhipu / Z.AI coding-plan has already supported the new generation of models, but DSH’s built-in catalog hasn’t kept up yet, GLM-5.3 might not be visible in the model selector.

auuduu/dsh-zai-coding-models is a DSH bridge plugin. It does not replace DSH’s provider configuration, nor does it modify API keys, endpoints, or other providers; it simply supplements GLM-5.3 on the standard zai-coding-cn provider route while preserving the original six bundled models. Below is an introduction to its features, installation methods, manual configuration, and limitations.

Plugin Positioning¶

Repository: https://github.com/auuduu/dsh-zai-coding-models
License: MIT
Node Requirement: package.json engines requires Node >=18
Target Audience: Users who have configured ZAI_CODING_CN_API_KEY, but DSH’s bundled pi-ai catalog has not yet exposed GLM-5.3
Ecosystem Position: This is a third-party bridge package within the DSH plugin ecosystem, not an official DeepSeek / Huanshan app store.

Core Features¶

Expose the latest Zhipu / Z.AI coding-plan models on the standard zai-coding-cn provider route, currently GLM-5.3.
Preserve the six bundled zai-coding-cn models: glm-4.5-air, glm-4.7, glm-5-turbo, glm-5.1, glm-5.2, glm-5v-turbo.
Add glm-5.3: text-only capability, 1M context, 128K max output, thinking levels Low / Medium / High / Max.
Set compatibility flags required for the endpoint: thinkingFormat: zai, supportsReasoningEffort: true.
Does not change API keys, endpoints, or other providers.
Provides a manual recipe that does not require installing the plugin, which can be written to ~/.dsh/settings.yaml.

Installation and Activation¶

First, confirm the profile you wish to modify. The following commands use the web profile; in actual use, you can replace web with tui, headless, or your own profile.

dsh plugin --profile web add dsh-zai-coding-models                  # from npm
dsh plugin --profile web add github:auuduu/dsh-zai-coding-models   # from GitHub

Restart DSH once after installation. This is because plugin bundles are read during boot, not immediately after installation.

After restarting, the ZAI_CODING_CN_API_KEY saved via the DSH Models page will be picked up as-is. GLM-5.3 will appear in the model selector of the existing Z.AI Coding CN provider.

If you confirm that you no longer need this plugin, the removal command is as follows:

dsh plugin --profile web remove dsh-zai-coding-models

Configuration Effect¶

The plugin declares a layer via dsh.bundle.patch, targeting the standard llm-pi-ai composition row and providing a base provider profile for zai-coding-cn. Simplified, what it does is: preserve the original model list and supplement it with the configuration for glm-5.3.

- id: llm-pi-ai
  config:
    providers:
      zai-coding-cn:
        models:
          - id: glm-4.5-air
          - id: glm-4.7
          - id: glm-5-turbo
          - id: glm-5.1
          - id: glm-5.2
          - id: glm-5v-turbo
          - id: glm-5.3
            name: GLM-5.3
            contextWindow: 1000000
            maxTokens: 131072
            reasoningEfforts:
              low: low
              medium: high
              high: high
              max: max
            compat:
              thinkingFormat: zai
              supportsReasoningEffort: true

Where:

contextWindow: 1000000 corresponds to 1M context.
maxTokens: 131072 corresponds to 128K max output.
reasoningEfforts provides Low / Medium / High / Max thinking levels; GLM-5.3 cannot disable thinking, Low is the lightest setting.
supportsReasoningEffort: true is required. Without it, the effort parameter will be silently dropped.

Manual Recipe: Without Installing the Plugin¶

If you don’t want to install the plugin, you can also write the same configuration directly to ~/.dsh/settings.yaml.

llm-pi-ai:
  providers:
    zai-coding-cn:
      apiKeyEnv: ZAI_CODING_CN_API_KEY   # keep whatever you have
      models:
        - id: glm-4.5-air
        - id: glm-4.7
        - id: glm-5-turbo
        - id: glm-5.1
        - id: glm-5.2
        - id: glm-5v-turbo
        - id: glm-5.3
          name: GLM-5.3
          contextWindow: 1000000
          maxTokens: 131072
          reasoningEfforts:
            low: low
            medium: high
            high: high
            max: max
          compat:
            thinkingFormat: zai
            supportsReasoningEffort: true

If you have already declared models for zai-coding-cn in ~/.dsh/settings.yaml, the user layer will override the plugin layer. It is recommended to keep only one source to avoid duplicate declarations of the same set of models by both the plugin and your own configuration.

Applicable Scenarios and Notes¶

Suitable for:

Users who have configured ZAI_CODING_CN_API_KEY, but DSH’s bundled catalog does not currently show GLM-5.3.
Need to use GLM-5.3’s 1M context, 128K max output, and thinking levels on the zai-coding-cn route.
Do not want to modify the DSH main workflow for a new model and only want to use the plugin bridge to fill in the model list.

Notes:

The plugin runs with the permissions of the current dsh process. It is recommended to check the GitHub source code, MIT license, and dependencies before installation to ensure behavior matches expectations.
The models list will replace the catalog for that route; this is DSH semantics. The list is fixed according to the plugin release.
When pi-ai releases glm-5.4 in the future, this plugin needs to release a new version to include the new model.
The plugin patch replaces the entire config of the llm-pi-ai row. The standard row currently has no config, but if DSH places config there in the future, the plugin may need a bump.
The thinking for GLM-5.3 cannot be disabled; Low is the lightest setting.
supportsReasoningEffort: true must exist, otherwise the effort parameter will be silently dropped.
package.json engines require Node >=18.

Conclusion¶

The value of dsh-zai-coding-models lies in a short window period: allowing the zai-coding-cn route to use GLM-5.3 before the bundled pi-ai catalog catches up. If DSH / pi-ai’s built-in directory already contains the target model later on, you can remove the plugin and revert to the catalog.

GitHub: https://github.com/auuduu/dsh-zai-coding-models

dsh-zai-coding-models: Early integration of GLM-5.3 on the `zai-coding-cn` track