Promote to production

You've got vendor-screen v1 and v2 in sandbox. Now we'll get the workflow to prod, activate it, and set up an API key for future automated deploys.

Step 1 — deploy to prod

culvii deploy vendor-screen.ts --env prod --workspace procurement

(Adjust the workspace name to whatever makes sense for your team. The Console will ask to create it if it doesn't exist.)

You'll see:

✓ Published vendor-screen v1 to prod

Note: it's v1 in prod, not v2. Versions are per-environment. Sandbox went v1, v2 independently of prod.

Step 2 — activate

In the Console, navigate to prod → procurement → workflows → vendor-screen. Click Activate.

The Console asks you to confirm; it shows you the workflow's input schema, output schema, and any triggers. (For our tutorial workflow there are no triggers, so this is mostly a no-op.) Click Activate.

The workflow is now Live in prod. The Console marks it with a "Live" badge. If you had cron triggers or webhooks attached, they'd be armed now.

Run it once manually from the Console with { "vendorId": "V001" } to confirm it works in prod.

Step 3 — create API keys for CI

So far you've been deploying as yourself, using the OAuth session created by culvii login. That works for one-off human-driven deploys. For anything automated — CI on merge to main, scheduled redeploys, integration tests — you'll need an API key.

Create a key for sandbox first:

culvii key create --env sandbox --scope developer --name "github-actions-sandbox"

The CLI prints the secret once:

Created API key:
  Fingerprint: ck_sandbox_a3f9...8c2d (last 4)
  Secret:      ck_sandbox_a3f9b8e6c2d4f7a1...

⚠ This is the only time the secret will be shown. Store it somewhere safe.

Repeat for prod:

culvii key create --env prod --scope developer --name "github-actions-prod"

In your CI provider (GitHub Actions, GitLab CI, whatever you use), store these as secrets:

• CULVII_API_KEY_SANDBOX = ck_sandbox_…
• CULVII_API_KEY_PROD = ck_prod_…

A minimal GitHub Actions example:

.github/workflows/deploy.yml

name: Deploy

on:
  push:
    branches: [main]
  workflow_dispatch:
    inputs:
      env:
        type: choice
        options: [sandbox, prod]

jobs:
  deploy-sandbox:
    if: github.event_name == 'push'
    runs-on: ubuntu-latest
    environment: sandbox
    steps:
      - uses: actions/checkout@v4
      - uses: actions/setup-node@v4
        with: { node-version: 20 }
      - run: npm install -g @culvii/cli
      - run: culvii deploy vendor-screen.ts --env sandbox --workspace tutorial --yes
        env:
          CULVII_API_KEY: ${{ secrets.CULVII_API_KEY_SANDBOX }}

  deploy-prod:
    if: github.event_name == 'workflow_dispatch' && github.event.inputs.env == 'prod'
    runs-on: ubuntu-latest
    environment: prod
    steps:
      - uses: actions/checkout@v4
      - uses: actions/setup-node@v4
        with: { node-version: 20 }
      - run: npm install -g @culvii/cli
      - run: culvii deploy vendor-screen.ts --env prod --workspace procurement --yes
        env:
          CULVII_API_KEY: ${{ secrets.CULVII_API_KEY_PROD }}

The --yes flag suppresses interactive prompts (workspace creation, etc.) — required for unattended runs.

This shape — sandbox on every merge, prod via manual dispatch — is the most common starting point. As you build confidence, you can wire prod deploys to a release branch, a tag, or a manual approval gate in your CI provider.

You're done

That's the full developer loop:

1. Write TypeScript with @culvii/kit.
2. Iterate with culvii dev against your dev environment.
3. Deploy to sandbox with culvii deploy --env sandbox.
4. Deploy and activate in prod.
5. Hand future deploys to CI via API keys.

Where to go next

• How-to guides — recipes for specific tasks: connecting MCP, adding a HITL gate, debugging with culvii logs --follow, real internal-API integrations.
• Concepts — if any of the model felt fuzzy, this is where it gets explained in depth.
• SDK reference — the precise interfaces of agent, tool, model, workflow.
• Operator's Guide — if you'll be the one approving HITL gates and reviewing the audit log, this is your home.

If anything broke along the way, email [TODO support@culvii.com] with your run ID. We read every message.