Your spreadsheet stays out of our database entirely on every paid plan. The free plan is the one exception: its conversations are logged and read by our team, as set out below. One model is a second exception on every plan: pick Meta Muse Spark 1.3 on Meta's Contributor tier and your prompts, replies and workbook values go to Meta, and Meta trains on them. That Contributor tier is the only model the free plan and the paid Lite plan run on. The standard Meta tier, on Pro, is not used by Meta for training. Here is the exact data path, including the cases where a request passes through our infrastructure, and what happens to it when it does.
Included usage on every plan passes through an EBITDAI relay to whichever included model you pick: Kimi k3, DeepSeek V4.1 Flash, DeepSeek V4 Pro, Gemini 3.8 Flash and Meta Muse Spark 1.3 on Pro, and Meta Muse Spark 1.3 on Meta's Contributor tier as the only model on the Free plan and Lite, which run the same tier. The relay forwards prompt and reply without logging or storing their content, and records only token counts for billing your included usage. With a key of your own the path depends on the provider: DeepSeek and Moonshot (pay-as-you-go Kimi) accept browser calls, so the add-in calls them directly from your machine and nothing touches our infrastructure, while Kimi Code and Gemini cannot be called from a browser, so those requests pass through a stateless relay that stores no prompt, workbook content or key; own keys are a Pro feature. Free plan: the relay also keeps a log of each conversation (prompt, reply, the analyst's actions and the workbook context sent to the model) for product review; see the table below. On any plan, picking Meta Muse Spark 1.3 on Meta's Contributor tier sends your prompt, the reply and the workbook values in it to Meta, and Meta trains on them; the free plan and Lite run that tier only. Pro also offers Meta Muse Spark 1.3 on Meta's standard tier, which Meta does not use for training.
When you run on your plan's included usage you have no key at all: EBITDAI's own provider credential lives server-side and never reaches your browser. Bring your own key instead (on Pro, your own key for any of the four providers) and it is stored in your browser's local storage on your device, never written to an EBITDAI database, and on relay paths forwarded per-request and then discarded: never logged, never retained. Both Meta Muse Spark 1.3 tiers are a different case: they run on EBITDAI's own Meta credential rather than a key of yours, and on the Contributor tier everything you send goes to Meta, which trains on it.
EBITDAI's backend (Supabase) holds account and billing records, not your spreadsheets. The full list is below, and EBITDAI does not train models on your data. Meta does on one tier: pick Meta Muse Spark 1.3 on Meta's Contributor tier, the only model on the free plan and on the paid Lite plan and an option on Pro, and everything you send with it goes to Meta for training. Meta's standard tier, on Pro, is not used for training.
Five categories of data, all in service of running and improving the product. On paid plans none of it is your workbook; on the free plan, the conversation log includes the workbook context you send to the model.
| Data | Why we store it |
|---|---|
| Account email | So you can sign in and we can reach you about your account. Managed through Clerk. |
| Subscription status | Whether you're on the free plan or a paid plan, synced from Stripe. We never see card numbers. |
| Usage counts | How many AI actions you've used, to administer the free plan's weekly requests and your plan. |
| Analytics events | Product analytics (page views, feature usage) to understand what to build next. |
| Free-plan conversations | Free plan only: your prompts, the analyst's replies, the actions it takes in the workbook and the workbook context sent to the model, kept for 90 days and read by the EBITDAI team to learn what works and what frustrates. The paid plans, Lite and Pro, are never logged by EBITDAI. Free-plan calls, like every call on the paid Lite plan, run on Meta Muse Spark 1.3 on Meta's Contributor tier, so Meta receives that same content and trains on it. Pro can also select Meta's standard tier, which is not used for training. |
EBITDAI uses a small number of well-known providers, each with a narrow role.
Authentication and user management: handles sign-in and sessions.
Payment processing: card details go to Stripe directly; EBITDAI never sees card numbers.
Backend database: stores the account, subscription, usage, and analytics records listed above.
Serves Meta Muse Spark 1.3 in two tiers, and which one you pick decides whether Meta trains on your workbook. On Meta's Contributor tier, shown in the pane as 'trains on your data', every prompt, reply and workbook value you send goes to Meta and Meta uses it to train its models; that is the only model on the Free plan and on the paid Lite plan, and an option on Pro. Meta's standard tier, shown as 'not used for training', is selectable on Pro, and Meta's published pricing terms say Meta does not use what you send there to train its models. Neither Meta option is ever the default.
Serves Gemini 3.8 Flash on the included usage that comes with Pro, and any Gemini key you connect yourself. Google processes those requests under the Gemini API terms.
Serves DeepSeek V4.1 Flash and DeepSeek V4 Pro, incorporated in the People's Republic of China and processing there. Both are included models on Pro, which can also connect your own DeepSeek key, called straight from your browser.
Serves Kimi k3 and Kimi Code, incorporated in Singapore. Kimi k3 is one of the included models on Pro; Pro also connects your own Kimi Code key (through EBITDAI's stateless relay) or your own Moonshot pay-as-you-go key (called straight from your browser). Neither the Free plan nor Lite reaches Moonshot.
Which provider receives a request follows the model you pick, not your plan. DeepSeek V4.1 Flash and DeepSeek V4 Pro are served by Hangzhou DeepSeek Artificial Intelligence Co., Ltd., incorporated in the People's Republic of China and processing there. Kimi k3 is served by Moonshot AI PTE. LTD., incorporated in Singapore. Gemini 3.8 Flash is served by Google LLC and Meta Muse Spark 1.3 by Meta Platforms, Inc., both incorporated in the United States. Connecting a key of your own (your own Kimi Code, Moonshot, DeepSeek or Gemini key on Pro) gives you a direct relationship with that provider, which then processes your prompts under its own terms. Meta Muse Spark 1.3 is served by Meta Platforms in the United States in two tiers: on Meta's Contributor tier, which is the only model for the free plan and Lite and an option on Pro, Meta trains on the prompts, replies and workbook content you send with it, while the standard tier on Pro is not used by Meta for training under its published pricing terms. Choose the model whose jurisdiction and terms fit your requirements.
The complete disclosures, including data retention periods and your rights, live in the Privacy Policy and Terms of Service. Security questions? Email joeorsi@ebitdai.co.