Xpert Plugin: MiniMax

Overview

@xpert-ai/plugin-minimax connects MiniMax AI models to the XpertAI platform. The plugin integrates MiniMax's OpenAI-compatible API so XpertAI agents can leverage MiniMax's large language models, text embeddings, and text-to-speech capabilities.

Core Features

• Ships MiniMaxModule, which registers the NestJS provider strategy, lifecycle hooks, and configuration schema required by the plugin runtime.
• Implements MiniMaxLargeLanguageModel, a LangChain-powered adapter built on ChatOAICompatReasoningModel that supports streaming chat completions, function calling, and token accounting callbacks for agent telemetry.
• Provides MiniMaxTextEmbeddingModel, a custom implementation that handles MiniMax's specific embedding API format with support for document and query embedding types.
• Exposes MiniMaxTTSModel, which supports streaming text-to-speech synthesis with multiple voice options and audio formats.
• Shares a console-ready manifest.yaml that drives the XpertAI UI forms (icons, help links, credential prompts) for quick operator onboarding.

Installation

npm install @xpert-ai/plugin-minimax

Peer Dependencies: Ensure your host service also provides @xpert-ai/plugin-sdk, @nestjs/common, @nestjs/config, @metad/contracts, @langchain/core, @langchain/openai, zod, and tslib. Refer to package.json for exact versions.

Enabling in XpertAI

1. Add the plugin to the service dependencies so Node.js can resolve @xpert-ai/plugin-minimax.
2. Declare the plugin before bootstrapping the XpertAI server:

   PLUGINS=@xpert-ai/plugin-minimax

3. In the XpertAI admin console (or config file), create a model provider pointing to minimax, then add individual models that map to the specific MiniMax model variants you want to use.

Credentials & Model Configuration

The plugin schema backs the form fields you see in the console:

| Field | Description | | --- | --- | | api_key | Required. Your MiniMax API Key from platform.minimaxi.com. | | group_id | Required. Your MiniMax Group ID from platform.minimaxi.com. | | base_url | Optional. Base URL for API requests (defaults to https://api.minimaxi.com). Useful for custom endpoints or proxy configurations. |

During validation, the plugin checks that both api_key and group_id are provided and validates the base URL format if specified.

Supported Models

Large Language Models (LLM)

• MiniMax-M2.5 - Latest flagship model, excels at coding and agentic tasks (196K context)
• MiniMax-M2.5-highspeed - High-speed variant of M2.5 with 100 tokens/s throughput
• MiniMax-M2.1 - Enhanced reasoning model with 230B parameters (196K context)
• MiniMax-M2.1-highspeed - High-speed variant of M2.1
• MiniMax-M2 - MiniMax M2 model with agentic capabilities (196K context)
• M2-her - Specialized for role-playing and multi-turn conversations (65K context)
• minimax-m1 - MiniMax M1 open-source model (1M context)

Text Embedding Models

• embo-01 - MiniMax embedding model

Text-to-Speech Models

• speech-2.8-hd - Latest HD TTS with 40 languages and 7 emotional variants
• speech-2.8-turbo - Latest turbo TTS with 40 languages and 7 emotional variants
• speech-2.6-hd - HD voice generation supporting 40 languages
• speech-2.6-turbo - Low-latency voice generation supporting 40 languages
• speech-02-hd - HD Speech-02 model with 24 languages
• speech-02-turbo - Turbo Speech-02 model with 24 languages

Model Capabilities

• Conversational Models: MiniMaxLargeLanguageModel merges provider credentials with per-model overrides, enables streaming, and registers token usage callbacks so agent telemetry stays accurate.
• Embedding Models: MiniMaxTextEmbeddingModel implements custom embedding logic to handle MiniMax's specific API format, supporting both document and query embedding types.
• TTS Models: MiniMaxTTSModel supports streaming text-to-speech synthesis with configurable voice settings, audio formats (mp3, wav, opus, aac, flac), and playback speed.

Development & Debugging

From the repo root, run Nx commands for this package:

cd xpertai
npx nx build models/minimax
npx nx lint models/minimax

Artifacts land in xpertai/models/minimax/dist. Jest settings live in jest.config.ts.

License

This plugin is distributed under the MIT License located at the repository root.