본문으로 건너뛰기

Creating Projects

Learn how to create and manage FluxLoop projects in VSCode.

Prerequisites​

  1. FluxLoop CLI and SDK installed

    pip install fluxloop-cli fluxloop

  2. FluxLoop VSCode Extension installed

    • Open VSCode Extensions (Cmd+Shift+X)
    • Search for "FluxLoop"
    • Click Install

Creating Your First Project​

Best for most users—FluxLoop reuses your open workspace as the agent source, auto-detects its environment, and places the FluxLoop configs in a shared root (~/FluxLoopProjects by default).

  1. Launch the wizard
    • Projects view → Create New Project…
    • or Command Palette → FluxLoop: Create FluxLoop Project
  2. Choose flow → Default (Recommended)
  3. Name your project (preview shows target location, e.g. ~/FluxLoopProjects/support-agent)
  4. Select environment
    • Wizard lists interpreters discovered via VS Code Python extension, .venv, Poetry, Conda, pyenv, uv, etc.
    • Options include:
      • ⭐ Detected workspace venv (preferred)
      • 📂 Choose another environment… to browse anywhere
      • ⚠️ Use system Python (fallback only)
    • If required packages (fluxloop-cli, fluxloop, fluxloop-mcp) are missing, a dialog offers:
      • Install automatically (runs pip inside chosen env)
      • Install manually (opens terminal with instructions)
      • Select different environment
  5. Include sample agent? (toggle)
  6. Finish
    • Wizard runs fluxloop init project inside the shared root
    • Sets fluxloop.projectRoot in workspace settings (change anytime under Settings → FluxLoop → Project Root)
    • Registers the project and makes it active in VSCode

Method 2: Custom Flow (Advanced)​

Use when you need full control—e.g., mono-repos, remote folders, or keeping configs beside your source tree.

  1. Run FluxLoop: Create FluxLoop Project
  2. Choose Custom (Advanced)
  3. Pick a destination folder for the FluxLoop project (can be inside the repo)
  4. Enter project name (created inside the folder you picked)
  5. Select environment:
    • Create/select a .venv within your repo, or browse to any interpreter/Conda env
    • Same package installation dialog appears if dependencies are missing
  6. Include sample agent (optional)
  7. Finish—FluxLoop writes configs exactly where you pointed and saves per-folder settings:
    • fluxloop.targetSourceRoot
    • fluxloop.executionMode
    • fluxloop.pythonPath / fluxloop.mcpCommandPath if custom executables were chosen

Project Structure​

After creation, your FluxLoop project (default flow) lives under the shared root:

~/FluxLoopProjects/my-chatbot/        # Default flow result (configurable)
├── configs/
│   ├── project.yaml       # Project metadata
│   ├── input.yaml         # Input generation
│   ├── simulation.yaml    # Experiment config
│   └── evaluation.yaml    # Evaluators
├── .env                   # Environment variables
├── examples/
│   └── simple_agent.py    # Sample agent
├── experiments/           # Results (created on run)
├── inputs/                # Generated inputs
└── recordings/            # Recorded args (optional)

Custom flow places the project wherever you choose, but the directory structure is identical.

FluxLoop also updates .vscode/settings.json with:

{
  "fluxloop.executionMode": "auto",
  "fluxloop.targetSourceRoot": "/path/to/your/workspace",
  "fluxloop.projectRoot": "~/FluxLoopProjects"
}

Adjust these in Settings UI or directly in the JSON file.

Configuring Your Project​

1. Update Project Metadata​

Click on Projects → Your Project → Configure

Edit configs/project.yaml:

project:
  name: my-chatbot
  version: "1.0.0"
  description: "Customer support chatbot"

2. Add Your Agent Code​

Replace the example agent:

# examples/chatbot.py
import fluxloop

@fluxloop.agent()
def run(user_message: str) -> str:
    """Your chatbot logic"""
    response = generate_response(user_message)
    return response

Update configs/simulation.yaml:

runner:
  target: "examples.chatbot:run"

3. Define Personas and Inputs​

Click Inputs → Base Inputs → Edit

Add to configs/input.yaml:

personas:
  - name: frustrated_customer
    description: Customer with unresolved issue
  
  - name: new_customer
    description: First-time user

base_inputs:
  - input: "I've been waiting for 3 days!"
    persona: frustrated_customer
  
  - input: "How do I create an account?"
    persona: new_customer

Managing Multiple Projects​

Add Existing Project​

If you already have a FluxLoop project (Default or Custom flow):

  1. Command Palette → FluxLoop: Add Existing FluxLoop Project
  2. Select the project directory (the folder containing configs/)
  3. Project added to Projects view (shared root detection is optional)

Switch Between Projects​

Click on any project in the Projects view to make it active.

The status bar shows the active project:

FluxLoop: my-chatbot

Remove Project​

Right-click project → Remove Project

This only removes from VSCode, doesn't delete files.

Project Status Indicators​

The Projects view shows status for each project:

  • ✅ Green check: Configuration valid
  • ⚠️ Yellow warning: Configuration issues
  • ❌ Red X: Missing required files

Click on the status icon to see details.

Best Practices​

1. One Project Per Agent​

Create separate projects for different agents:

fluxloop/
├── customer-support/
├── sales-assistant/
└── technical-qa/

2. Version Control​

Initialize git in your project:

cd fluxloop/my-chatbot
git init
git add configs/ examples/
git commit -m "Initial project setup"

Add .gitignore:

.env
experiments/
inputs/generated.yaml
recordings/

3. Environment Variables​

Store secrets in .env (never commit):

OPENAI_API_KEY=sk-your-key
ANTHROPIC_API_KEY=sk-ant-your-key

4. Start Simple​

Begin with:

  • 3-5 personas
  • 5-10 base inputs
  • 10 iterations per input

Expand as you validate the setup.

Troubleshooting​

"CLI Not Found"​

Install FluxLoop CLI:

pip install fluxloop-cli

Check status in Status view.

"Invalid Configuration"​

Click the warning icon to see specific issues:

  • Missing required fields
  • Invalid YAML syntax
  • File path errors

"Project Not Loading"​

Ensure your project has:

  • configs/ directory
  • Required YAML files
  • Valid .env file

Next Steps​