LLM Clients Overview

MiroFlow is model-agnostic — plug in any LLM and get better agent performance through smart rollback, iterative reasoning, and optimized tool orchestration. Switch models in one line of YAML. Same tools, same prompts, same environment.

Available Clients

Client	Provider	Model	Benchmark Status	Environment Variables
`MiroThinkerSGLangClient`	SGLang	MiroThinker series	GAIA, HLE, BrowseComp, xBench-DS, FutureX	`OAI_MIROTHINKER_API_KEY`, `OAI_MIROTHINKER_BASE_URL`
`ClaudeAnthropicClient`	Anthropic Direct	claude-3-7-sonnet	GAIA	`ANTHROPIC_API_KEY`, `ANTHROPIC_BASE_URL`
`GPT5OpenAIClient`	OpenAI	gpt-5	—	`OPENAI_API_KEY`, `OPENAI_BASE_URL`
`GPTOpenAIClient`	OpenAI	gpt-4o, gpt-4o-mini	—	`OPENAI_API_KEY`, `OPENAI_BASE_URL`
`ClaudeOpenRouterClient`	OpenRouter	anthropic/claude-3.7-sonnet, and other supported models	—	`OPENROUTER_API_KEY`, `OPENROUTER_BASE_URL`
`OpenRouterClient`	OpenRouter	Any model on OpenRouter	—	`OPENROUTER_API_KEY`, `OPENROUTER_BASE_URL`
`OpenAIClient`	OpenAI-Compatible	Any OpenAI-compatible model	GAIA Text-Only (Kimi K2.5)	`OPENAI_API_KEY`, `OPENAI_BASE_URL`

Basic Configuration

Agent Configuration

main_agent:
  llm:
    _base_: config/llm/base_mirothinker.yaml   # or base_openai.yaml
    provider_class: "MiroThinkerSGLangClient"
    model_name: "mirothinker-v1.5"

LLM Base Configs

Pre-configured base configurations are available in config/llm/:

Config File	Provider	Description
`base_mirothinker.yaml`	SGLang	MiroThinker model via SGLang
`base_openai.yaml`	OpenAI	GPT models via OpenAI API
`base_kimi_k25.yaml`	OpenAI-Compatible	Kimi K2.5 model

Quick Setup

Set relevant environment variables for your chosen provider
Update your YAML config file with the appropriate client and base config

Run:

bash scripts/test_single_task.sh \
  --config config/your_config.yaml \
  --task-question "Your task here"

Switch Models in One Line

Change provider_class and model_name in your YAML config — everything else stays the same:

# GPT-5
main_agent:
  llm:
    provider_class: GPT5OpenAIClient
    model_name: gpt-5

# Claude 3.7 Sonnet
main_agent:
  llm:
    provider_class: ClaudeAnthropicClient
    model_name: claude-3-7-sonnet-20250219

# MiroThinker (open-source, self-hosted)
main_agent:
  llm:
    provider_class: MiroThinkerSGLangClient
    model_name: mirothinker-v1.5

See the Model Comparison Leaderboard for cross-model benchmark results.

Documentation Info

Last Updated: March 2026 · Doc Contributor: Team @ MiroMind AI