Quickstart

1. Install & start (self-hosted / local)

git clone https://github.com/xxco/flightbox && cd flightbox
uv sync && uv run playwright install chromium
cp .env.example .env     # add ANTHROPIC_API_KEY / OPENAI_API_KEY / OPENROUTER_API_KEY
uv run flightbox serve   # http://localhost:8787

2. Define a suite

# suites/checkout.yaml
suite: checkout
runs_per_task: 3
agents: [browser-use-gpt, claude-cu]
tasks:
  - id: buy-tee
    start_url: "https://shop.example.com"
    goal: >
      Find a medium blue t-shirt under $30 and complete checkout as a guest
      with name Flight Box, zip 94016.
    oracle: {type: url-regex, pattern: "order-confirmed"}
    policies: ["never create an account", "never apply coupon codes"]
    budget: {max_steps: 30, max_usd: 1.50, max_minutes: 8}

3. Run it

uv run flightbox run suites/checkout.yaml
open runs/checkout/report/index.html     # scorecard -> replay synced to reasoning
uv run flightbox pdf runs/checkout       # executive PDF

Or through the API/SDK — see SDKs. Every run produces a black box: video, DOM replay (rrweb), Playwright trace, HAR, per-step screenshots, the agent's reasoning transcript, and friction telemetry.