Summary

• Add MiniMax as a first-class LLM provider alongside OpenAI, Gemini, Groq, Mistral, Ollama, Azure OpenAI, and T-Systems
• MiniMax models (M2.7, M2.7-highspeed, M2.5, M2.5-highspeed) are accessible via their OpenAI-compatible API using the existing llama-index-llms-openai-like dependency
• MiniMax supports function calling, enabling multi-agent workflows

Changes

Backend (Python):

• MiniMaxConfig model class with MINIMAX_API_KEY env var
• configured computed field recognizes minimax provider
• Settings patch (patch/backend/app/settings.py) adds init_minimax() using OpenAILike with temperature clamping to MiniMax's (0, 1] range
• Multi-agent support: MiniMax returns True for function-calling check

Frontend (TypeScript/React):

• minimax.ts provider schema with Zod validation
• minimax.tsx form component with API key input and model selector
• Registered in provider index, model config schema union, and UI dispatch

Tests:

• 22 unit tests covering config defaults, env mapping, configured check, multi-agent support, temperature clamping, and settings dispatch
• 3 integration tests for config roundtrip, env roundtrip, and OpenAILike LLM creation

Test plan

• All 22 unit tests pass
• All 3 integration tests pass (with MINIMAX_API_KEY)
• Existing test suite (test_agent_manager) passes without regressions
• Manual: select MiniMax in Admin UI, enter API key, verify chat works