Privacy Policy

This Privacy Policy explains what SPARK Engine (“we”, “us”) collects when you use the SPARK Engine desktop application and website, and — because this product is built around a local-first design — goes into more technical detail than most privacy policies about exactly what data leaves your machine, when, and in what form. If anything here conflicts with marketing copy elsewhere, this page is authoritative.

1. Account data

You sign in with Google, through AWS Cognito (our authentication provider). We receive your name, email address, and a stable account identifier from Google — we never see or store your Google password. This account data is stored in our AWS infrastructure (Mumbai, India region) and is used solely to identify your account, associate it with a subscription tier, and enforce usage quotas.

2. Billing data

Subscription payments are handled by Razorpay. We store only the metadata needed to know what plan you're on and whether your subscription is active — plan name, billing cycle, subscription status, and payment/order identifiers. We do not receive or store your card number, UPI ID, or bank details; those are handled entirely within Razorpay's own PCI-compliant systems.

3. Usage counters

To enforce the query limits attached to your tier, we store a small usage record per account — a count of queries run this month and today, and when those counters last reset. We do not store the content of your questions or query results as part of this record; it is purely numeric.

4. Your actual data: CSVs, spreadsheets, and databases

Any file or database you connect to SPARK Engine is imported into a local DuckDB instance running on your own machine. We do not upload, store, or have access to this data on any server we operate. It exists only on your device and is cleared when you disconnect the source (or, for CSV imports, when you remove the file).

5. What we send to AI providers — and what we don't

Turning a plain-English question into SQL, and reading an answer back to you, uses third-party AI models. Exactly what leaves your machine to make that happen depends on the privacy level you choose in Settings, and on your subscription tier. We designed this to be the one part of the product where the technical detail actually matters, so here it is in full:

Standard (default)

• Your question, your database's schema (table and column names, not their contents), and a handful of synthetic sample rows are sent to Groq so it can write accurate SQL.
• Before anything is sent, every value that looks like personal data — names, emails, phone numbers, addresses, and similar — is automatically detected and replaced with an opaque placeholder token (for example SPK_V17). Groq only ever sees the token, plus a short hint about its shape (“a date”, “an email address”) so it can still write a grammatically sensible sentence around it.
• The mapping from each token back to your real value is generated fresh for each query, lives only in your app's memory for the duration of that request, and is never written to disk, sent over the network, or logged.
• Ordinary non-personal values — counts, totals, product names, dates that aren't tied to a person — are not tokenized and may appear in the request in the clear.
• If voice output is enabled, the final answer is sent to a text-to-speech provider to be spoken aloud.

Strict

• Every value that came from your data — not just personal-looking ones — is tokenized before it leaves your machine.
• Cloud text-to-speech is turned off in this mode, because speaking the answer aloud would otherwise require sending the fully de-tokenized sentence to a second, separate provider.

Local / Offline (BLAZE tier and up)

• Nothing leaves your machine — not your question, not the schema, not sample rows. SQL generation runs entirely on-device using a locally-installed language model (node-llama-cpp), with no network call involved.
• This is the only mode where the common claim “your data never leaves your machine” is true in the fullest sense, including for the question you ask, not just the underlying data.

The safety check behind all of this

Independently of which level is active, every outbound request to an AI provider passes through a final check that inspects the exact text about to be sent and aborts the request if it finds a real, non-tokenized data value inside it. This exists so that a bug in the tokenization logic fails safe — as a blocked request — rather than as a silent leak.

6. Voice input and output

On tiers where voice is enabled, spoken questions are sent to Deepgram for speech-to-text transcription, and (outside Strict/Local modes) spoken answers are generated via a third-party text-to-speech provider. On the FREE through STORM tiers you supply your own Deepgram API key, so that audio is sent under your own account with that provider, subject to their terms. On THUNDER, we provide a managed key and that traffic runs through our own provider account instead.

7. Cookies and analytics

Our landing page does not use cookies, third-party analytics, or advertising trackers of any kind. It is a static page; we do not know who visits it beyond what your browser or ISP would ordinarily disclose to any website.

8. Where data is stored

Account, billing, and usage data described in sections 1–3 are stored on AWS infrastructure provisioned in the Mumbai (ap-south-1) region. We do not intentionally replicate this data outside India. Requests to third-party AI providers (section 5) are subject to those providers' own data-handling practices and may be processed outside India — this is limited to whatever a given privacy level permits leaving your device in the first place.

9. Data retention and deletion

We retain account, billing, and usage data for as long as your account is active, plus a reasonable period afterward for accounting and fraud-prevention purposes. You can request deletion of your account and associated data at any time by contacting us at hi@tanisheesh.in. Data that only ever existed on your own machine (your files, your local DuckDB databases) is deleted the same way any local file is — by removing it yourself or uninstalling the app.

10. Children's privacy

SPARK Engine is not directed at children and is not intended for use by anyone under 18. We do not knowingly collect personal data from children.

11. Your rights

Depending on where you live, you may have rights to access, correct, export, or delete the personal data we hold about you. To exercise any of these rights, contact us at hi@tanisheesh.in and we will respond within a reasonable time.

12. Changes to this policy

If we change what we collect or how we handle it — especially anything described in section 5 — we will update the “Last updated” date above and, for material changes, make a reasonable effort to notify you in-app.

13. Contact

Questions about this policy can be sent to hi@tanisheesh.in.