Xpert Plugin: Volcengine

Introduction

@xpert-ai/plugin-volcengine is a standard adapter for the XpertAI platform to access Volcengine (Doubao, etc.) large model services. The plugin connects to Volc Ark's OpenAI-compatible API, providing a unified entry point for agents to access capabilities such as conversation and function calling within a workflow.

Core Features

• Integrates Volcengine Ark platform's LLM/OpenAI-compatible interface, supporting key authentication, region selection, and multi-tenant configuration.
• Registers the VolcenginePlugin NestJS module, automatically wiring up model providers, lifecycle logging, and configuration validation logic for easy enablement in XpertAI applications.
• Provides VolcengineLargeLanguageModel, encapsulating Doubao-like conversational models with support for function calling, streaming output, token counting, and chain-of-thought exposure.

Supported Model Types

• Conversational Models: Function calling, tool calling, streaming output, and chain-of-thought reasoning.

Installation

npm install @xpert-ai/plugin-volcengine

Peer Dependencies: The host project must also provide libraries such as @xpert-ai/plugin-sdk, @nestjs/common, @metad/contracts, @langchain/openai, lodash-es, chalk, and zod. Please refer to package.json for specific versions.

Enabling in XpertAI

1. Install the plugin in the project where the XpertAI service runs, and ensure Node.js can resolve the package.
2. Before starting the service, declare the plugin via environment variables:

   PLUGINS=@xpert-ai/plugin-volcengine

3. In the XpertAI console or configuration file, add a new model provider, select volcengine, and fill in the corresponding Ark/Doubao model configuration.

Configuration

The configuration form is defined by volcengine.yaml, covering common Ark model deployment scenarios:

| Field | Description | | --- | --- | | api_key | Required. Access key for Volcengine Ark platform (temporary token generated by AK/SK or long-term token). | | endpoint_url | Required. Base URL for Volc Ark's OpenAI-compatible API, e.g., https://ark.cn-beijing.volces.com/api/v3. | | endpoint_model_name | Optional. If the server model name differs from the logical model name in XpertAI, override it here. | | mode | Choose chat or completion, corresponding to different Ark inference channels. | | context_size / max_tokens_to_sample | Controls context window and generation length; recommended to match Doubao model specs. | | agent_thought_support, function_calling_type, stream_function_calling, vision_support | Configure model capability tags for frontend display and agent strategy selection. | | region / workspace_id | Optional. Additional info for multi-region or multi-tenant Volc Ark deployments. | | stream_mode_delimiter | Custom delimiter for streaming output, useful for rendering and rich text processing. |

After saving, the plugin will call validateCredentials to send a minimal verification request to the Volc Ark API, ensuring the key and endpoint are correct.

Development & Debugging

In the repository root, enter xpertai/ and use Nx commands to build and test:

npx nx build @xpert-ai/plugin-volcengine
npx nx test @xpert-ai/plugin-volcengine

Build artifacts are output to the dist/ directory by default. Unit test configuration is in jest.config.ts; you can extend coverage as needed.

License

This project follows the AGPL-3.0 License in the repository root.