본문으로 건너뛰기

Project Setup

Create and configure your first FluxLoop project.

Initialize a New Project​

Create a new project with the CLI:

fluxloop init project --name my-agent
cd fluxloop/my-agent

This generates a complete project structure:

fluxloop/my-agent/
├── configs/
│   ├── project.yaml          # Project metadata
│   ├── input.yaml            # Input generation settings
│   ├── simulation.yaml       # Experiment configuration
│   └── evaluation.yaml       # Evaluation rules
├── .env                      # Environment variables
├── examples/
│   └── simple_agent.py       # Sample instrumented agent
├── experiments/              # Results output directory
├── inputs/                   # Generated inputs directory
└── recordings/               # Recorded arguments (optional)

Project Configuration Files​

configs/project.yaml​

Project-level metadata and collector settings:

project:
  name: my-agent
  version: "1.0.0"
  description: "My AI agent simulation project"

collector:
  enabled: false
  url: "http://localhost:8000"
  api_key: ""

configs/input.yaml​

Personas, base inputs, and input generation settings:

personas:
  - name: novice_user
    description: A user new to the system
    traits:
      - Asks basic questions
      - May not know terminology
  
  - name: expert_user
    description: An experienced power user
    traits:
      - Uses technical jargon
      - Expects advanced features

base_inputs:
  - input: "How do I get started?"
    expected_intent: help
    persona: novice_user
  
  - input: "What are the advanced configuration options?"
    expected_intent: configuration
    persona: expert_user

input_generation:
  mode: llm                    # or "deterministic"
  llm:
    provider: openai           # or "anthropic"
    model: gpt-4o-mini
    temperature: 0.7
  strategies:
    - rephrase
    - verbose
    - concise
    - error_prone

configs/simulation.yaml​

Experiment runtime configuration:

runner:
  target: "examples.simple_agent:run"   # Your agent function
  iterations: 10                         # Runs per input

replay_args:
  enabled: false
  recording_file: "recordings/args.jsonl"
  callable_providers: {}
  override_param_path: ""

output:
  directory: "experiments"
  format: jsonl

configs/evaluation.yaml​

Evaluator definitions (optional):

evaluators:
  - name: response_quality
    type: rule_based
    rules:
      - check: response_length_min
        value: 10
      - check: no_errors
        value: true

  - name: llm_judge
    type: llm
    model: gpt-4o-mini
    prompt: "Rate this response quality from 1-10"

Environment Variables​

The .env file stores sensitive configuration:

# LLM Provider API Keys
OPENAI_API_KEY=sk-your-key-here
ANTHROPIC_API_KEY=sk-ant-your-key-here

# Collector Configuration
FLUXLOOP_COLLECTOR_URL=http://localhost:8000
FLUXLOOP_COLLECTOR_API_KEY=your-api-key

# Recording Mode
FLUXLOOP_RECORD_ARGS=false

# Project Settings
FLUXLOOP_PROJECT_NAME=my-agent

Configure LLM Provider​

Set your API key for input generation:

# Using CLI helper
fluxloop config set-llm openai sk-your-api-key --model gpt-4o-mini

# Or manually edit .env
echo "OPENAI_API_KEY=sk-your-key" >> .env

Add Your Agent Code​

Replace the example agent with your actual agent:

# examples/my_agent.py
import fluxloop

@fluxloop.agent()
def run(input_text: str) -> str:
    """Your agent logic here"""
    # Process the input
    response = process_input(input_text)
    return response

Update configs/simulation.yaml to point to your agent:

runner:
  target: "examples.my_agent:run"
  iterations: 10

Define Personas and Base Inputs​

Edit configs/input.yaml to match your use case:

personas:
  - name: customer_support
    description: A customer seeking help
    traits:
      - Polite but frustrated
      - Wants quick resolution
  
  - name: technical_user
    description: A developer using the API
    traits:
      - Technical language
      - Needs documentation

base_inputs:
  - input: "My order hasn't arrived yet"
    expected_intent: order_status
    persona: customer_support
  
  - input: "What's the rate limit for the API?"
    expected_intent: api_docs
    persona: technical_user

Verify Configuration​

Check your project status:

fluxloop status

This shows:

  • ✅ CLI version
  • ✅ SDK installation
  • ✅ Project configuration
  • ✅ Environment variables
  • ✅ Recording mode status

Multiple Projects​

FluxLoop supports multiple projects:

# Create another project
fluxloop init project --name chatbot-v2
cd fluxloop/chatbot-v2

# Switch between projects
cd ../my-agent

Each project has independent configuration and results.

Best Practices​

  1. Version control your configs

    git add configs/
    git commit -m "Add project configuration"

  2. Keep .env out of version control

    echo ".env" >> .gitignore

  3. Use descriptive base inputs

    • Cover different personas
    • Include edge cases
    • Add expected outcomes

  4. Start small

    • Begin with 5-10 base inputs
    • Generate 50-100 variations
    • Run 10 iterations per input

  5. Iterate on configuration

    • Review results
    • Adjust personas
    • Refine generation strategies

Next Steps​