Getting started

Coming soon. Reflow V2 is in active development. This guide describes the launch workflow; follow the changelog for availability.

Ten minutes from signup to a verified PR.

1. Sign up and create a team

Sign in at app.reflow.io with Google or an email magic link, then create a team. Your team owns flows, runs, keys, and billing.

2. Add a model provider key

Settings → Provider Keys → Add. Paste your Anthropic or OpenAI key. Reflow encrypts it with KMS; runners decrypt it only for the duration of a run, and it never appears in logs. Reflow never bills you for tokens — your provider does.

3. Create an API token

Settings → API Tokens → Create, with the exec:runs scope. Copy it once — it’s hashed at rest and never shown again. Store it as the REFLOW_API_TOKEN secret in your GitHub repo.

4. Top up credits

Settings → Billing → Top up. Runs hold their worst-case cost up front and settle to actual. Optional auto-reload keeps the balance above your threshold. New accounts get $5 promo credit.

5. Add Reflow to your repo

your-repo/
  .reflow/
    config.yml
    flows/
      smoke.md

config.yml is three lines:

team: your-team-slug
default_provider_key: anthropic-prod

Write your first flow — see the flow format.

6. Wire the GitHub Action

.github/workflows/reflow.yml

on:
  pull_request:
jobs:
  reflow:
    runs-on: ubuntu-latest
    steps:
      - uses: actions/checkout@v4
      - uses: reflow-io/reflow-action@v1
        with:
          api-token: ${{ secrets.REFLOW_API_TOKEN }}
          target-url: ${{ steps.preview.outputs.url }}

Point target-url at whatever preview environment your PRs already get — Vercel, Netlify, or your own. Open a PR: Reflow runs every flow against the preview and posts a check with screenshots.

Working from an agent

Connect your coding agent to the MCP server and it can do all of the above discovery and execution itself:

claude mcp add reflow --transport http https://mcp.reflow.io \
  --header "Authorization: Bearer $REFLOW_TOKEN"