# Before adding a Gemini key

**Friends and pilot users:** create a key at https://aistudio.google.com/apikey. If AI Studio offers it without billing, use that (free tier is enough: the agent makes at most a few dozen small calls per run). If your region requires billing, follow steps 4 and 5 below (prepay the minimum, set a monthly cap) or ask Luka for the shared pilot key. The full checklist below is for the account owner.

Prepared 12 September 2026. This checklist is also linked in Model & settings. No key is needed for the mock tests. Do not paste an API key into this conversation.

1. Secure the Google owner account with two-step verification. Create a dedicated Google Cloud project named, for example, `WinkView pilot`; do not reuse a project with unrelated services or publicly distributed Firebase keys. In Google AI Studio, use Dashboard → Projects → Import projects if your project is not shown.
2. Open [AI Studio API keys](https://aistudio.google.com/api-keys) and create a new key in that project. Use the new authorization key type; Google says new AI Studio keys use this type by default. Check that it is restricted to the Generative Language API. Do not grant unrelated services access. [Official key instructions](https://ai.google.dev/gemini-api/docs/api-key).
3. Decide the billing and data-use arrangement before sending real DMs. Free quota may be useful for synthetic tests, but it is not a blanket privacy guarantee. Google's terms distinguish paid/unpaid processing and include regional provisions; API clients made available to users in the EEA, UK or Switzerland require Paid Services. Review the account's applicable terms and the creator's permission to process conversations. [Gemini terms](https://ai.google.dev/gemini-api/terms).
4. For a paid pilot, open AI Studio → Billing, select Prepay if available, and add only the minimum starting balance (currently $5). Leave automatic reload OFF. Verify the actual plan is Prepay; do not assume adding a card selected it. If your account only offers Postpay, stop here until you deliberately choose that exposure. [Billing instructions](https://ai.google.dev/gemini-api/docs/billing).
5. Open AI Studio → Spend → select this project → Monthly spend cap → Edit spend cap. Start at $5 if accepted by the account. Save and verify it remains selected. Also configure billing alerts for early warning. **Neither a prepaid balance nor the project cap is a precise real-time ceiling:** Google documents roughly ten minutes of accounting delay and possible overages. Alerts alone do not stop usage. [Project caps](https://ai.google.dev/gemini-api/docs/billing#project-spend-caps).
6. Check the project's available request/token quotas in AI Studio. Lower adjustable quotas to the pilot's needs, where the console permits it; the app itself permits six model calls per minute and one at a time. Do not request higher tiers for the pilot. [Rate limits](https://ai.google.dev/gemini-api/docs/rate-limits).
7. Once hosting has a stable outbound IP, use Cloud Console → APIs & Services → Credentials → the key → Application restrictions → IP addresses to restrict use to that server IP, if supported for your key. Do this only after the correct outbound IP is known; laptop IPs can change. Keep the Gemini-only API restriction. [Key restrictions](https://ai.google.dev/gemini-api/docs/api-key#restricting-and-securing-your-keys).
8. Add the key locally through the signed-in owner's Model & settings, or put `GEMINI_API_KEY` in the hosting secret store for a deployment. The production dashboard intentionally cannot edit provider secrets. After steps 1–7 are reviewed, set `PROVIDER_SETUP_CONFIRMED=true` on the backend. Keep `ALLOW_LIVE_POSTING=false` until the pilot is ready.
9. Test model connectivity, check the actual selected model/pricing, identify the exact Instagram creator, then deliberately enable live actions and authorise a short, low-cap pilot in the dashboard. Watch the browser. Check both the app's reserved allowance and Google's usage after its reporting delay before increasing anything.

If a key leaks: use Emergency stop in WinkView, immediately revoke/delete the compromised key in Google (service interruption is acceptable), inspect project usage, create a restricted replacement and investigate the cause before resuming. App limits cannot control calls made directly with a stolen provider key.

Current app limits are in USD: $0.05 maximum reservation per request, $0.20/session, $1/day and $10/calendar month for this one workspace. These are conservative allowances, not a guarantee about charges made outside this app. Google billing, hosting costs and the subscription price are separate.