Plain-language privacy. The lawyer version is over here.

Spark makes money one way: people pay a monthly or annual subscription for the app. That's it. We do not sell your contact list. We do not sell aggregated "insights" about your network. We do not license your data to AI training partners. We do not run ads.

The day we change any of that is the day the company has lost its reason to exist, and we know it.

Some "personal CRMs" ask for access to your Gmail and your iMessage. They scan everything you've ever written, build a profile, and call it "intelligence." That's an enormous amount of access in exchange for very little benefit.

Spark doesn't do that. The only messages we ever analyze are the messages you compose and send through Spark itself — and only to learn your writing voice so future AI drafts sound like you. Your broader email and text history is none of our business.

When you import your phone's contacts on day one, iOS shows you a system contact picker. You pick which contacts to import. Spark only ever sees the contacts you actively select.

We deliberately do not request the full Contacts permission that would give us access to your entire address book in the background. We never want that level of access, even if Apple ever made it easier to ask for.

When you hold the mic to talk to Spark, your speech is transcribed using Apple's on-device speech recognition. The audio never leaves your phone. The text we receive on the server is the same text you saw on your screen — nothing more.

The people you store in Spark didn't sign up to be in a CRM. They're in your contact list because they know you — that's the contract. So we're careful with their data the same way we're careful with yours: nothing leaves Spark except in service of features you've explicitly turned on, and nothing gets sold.

When you delete a contact, all the notes, dates, and AI-distilled facts about them are deleted with them. Including from our backups, within 30 days.

Spark uses third-party language models (currently Claude from Anthropic) to draft messages, refresh contact bios, and run the morning curation job. We have a data processing agreement with our model provider. They do not train models on your data. We do not retain raw prompt/response pairs beyond what's needed to debug a specific issue.

For Pro users, news grounding works by deduplicating topics across all Pro users into a single global search per topic — so if 200 people in Spark follow the Ravens, there's one Ravens search per day, not 200. Your contacts and your topics are never sent to the news API. Only the topic string.

Settings → Export Data downloads a complete JSON of everything Spark knows: contacts, notes, dates, the voice profile we've built, the topics we've extracted, the AI action log. The export is yours to keep whether or not you stay a customer.

If you cancel and delete your account, all of it is erased from our systems within 30 days. We do not keep a shadow copy "just in case."

We don't think you should have to trust an AI assistant on faith. So every autonomous thing Spark does — every bio refresh, every cross-attributed note, every inferred relationship — is logged in Settings → Spark activity. Tap a row, jump to the contact it touched, undo it if you don't like it. Within 24 hours, everything is reversible.