Skip to content

YAML Configuration Guide

MiroFlow uses a Hydra-based configuration system for customizing AI agents, tools, and benchmarks.

Configuration Structure

Configuration Directory
config/
├── agent_*.yaml                      # Agent configurations
├── agent_prompts/                    # Prompt classes
├── benchmark/                        # Benchmark settings
└── tool/                             # Tool configurations

Quick Start

Run Benchmarks 

# GAIA validation
uv run main.py common-benchmark \
  --config_file_name=agent_gaia-validation \
  output_dir="logs/gaia-validation/$(date +"%Y%m%d_%H%M")"

# GAIA text-only
uv run main.py common-benchmark \
  --config_file_name=agent_gaia-validation-text-only \
  output_dir="logs/gaia-validation-text-only/$(date +"%Y%m%d_%H%M")"

Single Task 

uv run main.py trace \
  --config_file_name=agent_quickstart_1 \
  --task="Your task here" \
  --task_file_name="data/file.xlsx"

Core Configuration

Basic Agent Setup

Basic Agent Configuration
defaults:
  - benchmark: gaia-validation
  - override hydra/job_logging: none
  - _self_

main_agent:
  prompt_class: MainAgentPromptBoxedAnswer
  llm:
    provider_class: "ClaudeOpenRouterClient"
    model_name: "anthropic/claude-3.7-sonnet"
    temperature: 0.3
    max_tokens: 32000
    openrouter_api_key: "${oc.env:OPENROUTER_API_KEY,???}"

  tool_config: []  # Tools for main agent
  max_turns: -1    # -1 = unlimited

sub_agents:
  agent-worker:
    prompt_class: SubAgentWorkerPrompt
    llm:
      provider_class: "ClaudeOpenRouterClient"
      model_name: "anthropic/claude-3.7-sonnet"
    tool_config:
      - tool-reading
      - tool-searching
    max_turns: -1

output_dir: logs/
data_dir: "${oc.env:DATA_DIR,data}"

LLM Providers

Available Providers

  • Claude: ClaudeOpenRouterClient, ClaudeAnthropicClient
  • OpenAI: GPTOpenAIClient
  • MiroThinker: MiroThinkerSGLangClient
  • Qwen: QwenSGLangClient
  • DeepSeek: DeepSeekNewAPIClient (limited support)

See LLM Clients Overview for details.

Available Tools

Tool Options

  • tool-reasoning: Enhanced reasoning capabilities
  • tool-searching: Web search and retrieval
  • tool-reading: Document processing
  • tool-code: Python code execution
  • tool-image-video: Visual content analysis
  • tool-audio: Audio processing
  • tool-browsing: Web browsing

See Tool Overview for configurations.

Advanced Features

GAIA Benchmark Configuration

GAIA-Optimized Setup
main_agent:
  prompt_class: MainAgentPrompt_GAIA
  tool_config:
    - tool-reasoning

  input_process:
    o3_hint: true              # Use O3 for task hints
  output_process:
    o3_final_answer: true      # Use O3 for answer extraction

sub_agents:
  agent-worker:
    tool_config:
      - tool-searching
      - tool-reading
      - tool-code
      - tool-image-video
      - tool-audio

Benchmark Settings

Benchmark Configuration
name: "your-benchmark"
data:
  data_dir: "${data_dir}/your-data"
execution:
  max_tasks: null      # null = no limit
  max_concurrent: 3    # Parallel tasks
  pass_at_k: 1         # Attempts per task

Environment Variables

Required .env Configuration
# LLM Providers
OPENROUTER_API_KEY="your_key"
ANTHROPIC_API_KEY="your_key"
OPENAI_API_KEY="your_key"

# Tools
SERPER_API_KEY="your_key"
JINA_API_KEY="your_key"
E2B_API_KEY="your_key"

# Optional
DATA_DIR="data/"
CHINESE_CONTEXT="false"

Key Parameters

Parameter Description Default
temperature LLM creativity (0.0-1.0) 0.3
max_tokens Response length limit 32000
max_turns Conversation turns (-1 = unlimited) -1
max_tool_calls_per_turn Tool calls per turn 10
max_concurrent Parallel benchmark tasks 3

Best Practices

Quick Tips

  • Start simple: Use agent_quickstart_1.yaml as a base
  • Tool selection: Choose tools based on your task requirements
  • API keys: Always use environment variables, never hardcode
  • Resource limits: Set max_concurrent and max_tokens appropriately
  • Development: Use higher temperature and unlimited max_turns for exploration

Documentation Info

Last Updated: September 2025 · Doc Contributor: Team @ MiroMind AI