dsh-glm-web: Zhipu Qingyan Web Login Session Integration with DeepSeek Harness

DeepSeek Harness (DSH) is a plugin-based agent runtime framework. lequait/dsh-glm-web is a model inference plugin that converts the web login state of Zhipu Qingyan (chatglm.cn) into a native DSH model provider. With this plugin, you can run agents directly in DSH using web credits without consuming Open Platform API quota.

Feature Overview¶

• Native Provider: Registers as a native DSH model provider and can be used directly in the model selector.
• Consumes Web Credits: Uses the web endpoint and does not occupy Open Platform API quota.
• Token Management: Supports both Refresh Token and Access Token modes, and supports automatic token refresh.
• Conversation Capabilities: Supports streaming conversations and tool calling.

Installation and Activation¶

Install it using DSH’s plugin management command:

dsh plugin --profile desktop add dsh-glm-web

After installation, restart DSH. After restarting, find “Qingyan Web · General” in the model selector.

Login State Configuration¶

The plugin requires a web login state to run. There are two configuration methods:

1. File Configuration: Write the token to ~/.dsh/storages/glm-web/token.txt.
2. DSH Settings: In the plugin configuration on the DSH settings page, find the dsh-glm-web plugin and fill in the refreshToken field.

How to obtain the token: Log in to chatglm.cn in a browser, inspect the Cookie via F12, and obtain chatglm_refresh_token (default validity: 180 days) or chatglm_token (valid for approximately 2 hours).

Models and Usage¶

• General Model: glm-web/chat
• Deep Thinking Model: glm-web/deep

This plugin supports streaming conversations and tool calling. Use the model IDs above in DSH prompts to perform inference and consume web credits.

Notes¶

• Unofficial Client: This project is not an official Zhipu implementation. It is intended for personal learning and research only, and commercial use is prohibited.
• Risk Assumption: Using private interfaces may trigger risk controls or violate terms of service.
• Concurrency Limits: A single account usually allows only one concurrent session.
• Token Consumption: Web sessions are deleted after use, and the full conversation history is resent each turn, so long sessions consume many credits.
• Protocol Risk: Web-side updates may invalidate the protocol.