glm-for-copilot: A BYOK Solution for Integrating Zhipu GLM Large Models into GitHub Copilot

Project Basic Information

• Original Author/Maintainer: KiwiGaze
• Source Platform: GitHub
• Original Link: https://github.com/KiwiGaze/glm-for-copilot
• Release Time: June 15, 2026

Core Points

The glm-for-copilot project enables developers to use Zhipu AI's GLM series large models (GLM-4.7/5/5.1/5.2/4.5 Air) in GitHub Copilot Chat, supporting Bring Your Own Key (BYOK), thinking mode, tool calling, and Agent mode. It aims to solve issues such as network latency, cost, data outbound concerns, and limited model choices caused by GitHub Copilot's default use of OpenAI models, providing developers with a more flexible, compliant, and cost-effective AI coding assistant solution.

As a popular AI coding assistant, GitHub Copilot uses OpenAI GPT models by default, but it has the following pain points for Chinese developers:

1. Network Latency: Overseas servers lead to high latency and unstable access in China;
2. Cost Issues: Copilot Pro subscription fees are an ongoing expense for individuals or small teams;
3. Data Outbound Concerns: Sensitive project code transmitted overseas may pose compliance risks;
4. Limited Model Choices: The official version only supports specific models, making it impossible to flexibly switch to preferred models.

The glm-for-copilot project was created to address these issues, providing a technical solution to integrate Zhipu GLM series models into Copilot.

Supported GLM Models

Zhipu AI's GLM series models perform well in tasks like code generation and mathematical reasoning. The project supports:

• GLM-4.7 (Flagship level, strong comprehensive capabilities)
• GLM-5/5.1/5.2 (New generation series)
• GLM-4.5 Air (Lightweight high-speed model)

These models can be accessed via Z.ai (Zhipu International Version) or Zhipu Open Platform.

Advantages of BYOK Mode

The project adopts the Bring Your Own Key (BYOK) mode, where users provide their own API keys. The advantages include:

• Cost Control: Pay based on actual usage, avoiding fixed subscription expenses;
• Flexible Switching: Switch between multiple models at any time to adapt to different task complexities;
• Data Sovereignty: Code is only sent to the API endpoint configured by the user, meeting the compliance requirement of data not leaving the country;
• Transparent Billing: Usage statistics are available through the Zhipu platform, allowing clear understanding of call costs.

1. Native Model Selector Integration: Seamlessly integrated with Copilot's native model selector, allowing direct selection of GLM models from the dropdown menu;
2. Thinking Mode: Supports deep thinking functionality, enabling multi-step reasoning for complex programming problems, suitable for tasks like algorithm design and architecture decisions;
3. Tool Calling: Integrates GLM's function calling capabilities with Copilot's tool ecosystem, such as code search, terminal command execution, and document query;
4. Agent Mode: AI autonomously plans and executes multi-step tasks, such as codebase refactoring suggestions, automatic test case generation, and creation of complete code files;
5. Dual API Access: Supports Coding Plan (code generation optimization API) and Standard API (general dialogue API).

Technical Implementation

The project acts as a model adaptation layer, implementing format conversion between Copilot and GLM API:

• Request Conversion: Intercepts Copilot's OpenAI format requests and encapsulates them into GLM API format;
• Streaming Response Processing: Converts GLM's streaming output into the SSE format expected by Copilot;
• Tool Calling Protocol: Maps Copilot's tool calling format to GLM's function calling format;
• Model Metadata: Provides model lists and capability statements to enable Copilot to correctly recognize GLM models.

Deployment Steps

1. Obtain GLM API Key: Register and create one on Z.ai or Zhipu Open Platform;
2. Deploy Adaptation Service: Run locally, deploy on a server, or use containerized deployment;
3. Configure Copilot: Modify the model endpoint to point to the adaptation service;
4. Verify Connection: Send a test message in Copilot Chat to confirm the response.

Developers with strong technical skills can complete the deployment within 30 minutes.

Use Cases

• Individual Developers: Reduce costs, stable access, and flexible model selection;
• Enterprise Teams: Meet data compliance requirements, unified usage management, and avoid multi-user subscriptions;
• Model Researchers: Evaluate GLM's code capabilities, compare model performance, and collect feedback to improve models.

Limitations and Notes

• Function Compatibility: Community projects may lag behind Copilot's new features;
• Stability Risk: The stability of self-built services depends on the deployment environment and maintenance;
• Technical Support: Dependent on the community, no official customer service;
• Compliance Usage: Need to comply with Zhipu API terms and Copilot service terms.

Future Outlook

• Support more new GLM models (e.g., GLM-6 series);
• Achieve more complete Copilot feature coverage;
• Provide simpler deployment solutions (e.g., one-click installation package);
• Integrate more domestic large models (e.g., Wenxin Yiyan, Tongyi Qianwen, etc.).