dsh-llm-volcengine

When integrating Volcengine Ark models into DeepSeek Harness, developers often encounter compatibility issues with Agent Plan and Coding Plan endpoints, such as unsupported thinking-effort controls or gateway rejections due to parameter mismatches. dsh-llm-volcengine is designed to resolve these challenges by providing a self-contained adapter that directly constructs OpenAI-compliant Provider objects. It ensures seamless operation across various models by correctly routing requests to specific endpoints—like using the OpenAI Responses path—and handling strict compatibility flags required by the Coding gateway.

The plugin's core capability lies in its support for both Agent Plan and Coding Plan scenarios, with verified configurations for a wide range of models including DeepSeek, GLM, Kimi, MiniMax, and Doubao-seed. It intelligently manages model-specific constraints, such as context windows and token limits, while supporting flexible credential management through environment variables or the DSH credential store. This ensures that developers can focus on building applications rather than debugging API integration quirks.

This plugin is ideal for developers using DeepSeek Harness who wish to leverage the power of Volcengine Ark's advanced reasoning models. Whether you are building complex Agents or specialized Coding solutions, this plugin offers a stable, verified bridge that simplifies the integration process and unlocks the full potential of Volcengine's high-performance models.