feat: add MiniMax provider support for voice agent LLM

[octo-patch]

Summary

• Add MiniMaxService LLM provider class that connects to MiniMax's OpenAI-compatible API
• Support MiniMax-M2.7 (default) and MiniMax-M2.7-highspeed models
• Add minimax as a new backend option in get_llm_service_from_config
• Add example config llm_configs/minimax.yaml for voice agent setup
• Add unit tests for MiniMaxService and the factory function

Changes

nemo/agents/voice_agent/pipecat/services/nemo/llm.py

• New MiniMaxService(OpenAILLMService) class:
  • Reads API key from MINIMAX_API_KEY environment variable (or explicit api_key)
  • Default base URL: https://api.minimax.io/v1 (overseas endpoint)
  • Temperature must be in range (0.0, 1.0] — MiniMax does not support temperature=0
• Updated get_llm_service_from_config to handle type: minimax
• Updated assertion to include "minimax" in supported backends

examples/voice_agent/server/server_configs/llm_configs/minimax.yaml

New config file for using MiniMax as the voice agent LLM backend.

examples/voice_agent/tests/test_minimax_llm.py

Unit tests covering instantiation, env-var key resolution, missing key error, default URLs, supported models, and factory function routing.

API Reference

• Chat (OpenAI Compatible): https://platform.minimax.io/docs/api-reference/text-openai-api

Usage

Set MINIMAX_API_KEY environment variable, then configure the voice agent server:

llm:
  type: minimax
  model: "MiniMax-M2.7"
  model_config: "./server_configs/llm_configs/minimax.yaml"

[pzelasko]

@stevehuang52 can you take a look?

[stevehuang52]

Does it work with MiniMax-M2.5 or other versions? We can probably drop this hardcoded SUPPORTED_MODELS if all MiniMax models support the same API.

[stevehuang52]

We can probably drop this test since the list of supported models may change frequently over time?