Privacy Policy
Last updated: 2026-10-30. Nodal helps people understand, document, and recover HSA, FSA, HRA, and ICHRA dollars. The public website, Chrome extension, mobile beta, and receipt workflows collect different data, and each surface is described below.
Website and early-access inquiries
When you submit the early-access form, we collect the work email, role, and optional context you provide. We use it to reply, route you to the right beta, employer-pilot, partner, or investor conversation, and keep a lightweight record of who asked to participate.
The form can send that information to Resend for email notification and Airtable for lead tracking. We do not sell this information, use it for advertising, or add tracking pixels to the site. We keep inquiry records while we manage beta and pilot relationships, and delete them on request unless a legal or security obligation requires us to keep them.
We count visits to the website with Vercel Web Analytics, run by our hosting provider. For each page view it records the page address without anything after a "#", the site that sent you (such as a search engine or an AI assistant), your approximate location, and your browser, operating system and device type. It sets no cookies and does not identify you or follow you across other sites: Vercel tells visits apart with a hash of the request that it discards after 24 hours. Text you type into the eligibility checker stays in the part of the address after the "#", so it is never sent. We use these counts only to see which pages people find and how they arrive.
What the extension can see
Chrome shows you a permission prompt at install: "Read and change all your data on the websites you visit." That's the install-time permission. It describes what the extension is technically able to observe. It does not describe what we transmit. The two are different and the difference matters.
The extension's content script extracts a small, fixed shape of product metadata from the page you're on: title, brand, price, category breadcrumbs, a query-stripped product path or slug when it helps identify the item, and (where present) the structured Schema.org product data the merchant publishes for search engines. That's it. We don't read your cookies, query strings, session tokens, shopping cart, payment information, or any other page content.
What leaves your device
Most eligibility decisions are made entirely on your device by a rules engine, a local product corpus, and Chrome's built-in on-device language model (Gemini Nano). Nothing about those decisions touches our servers.
When the on-device decision is uncertain, the extension sends the extracted product metadata
(the same ProductSignal shape described above) to our backend at
api.nodalhealth.ai. That request includes:
- The product metadata (title, brand, price, category, structured Schema.org product data).
-
The merchant identifier (a stable short string like
amazonorwalgreens). - A randomly-generated verdict ID for our own debugging.
- A per-install authentication token. Each installation of the extension is issued its own revocable, non-identifying token when it registers with our backend. The token is a random identifier plus a secret; the raw secret is never stored on our servers, and the token identifies the installation, not you.
It does not include: the full page URL, query string, cookies, anything you typed into the page, your shopping cart, payment information, or any account identifier. Like any internet request, the network layer carries IP-level routing information.
The mobile app works the same way, whether or not you sign in. When you check an item, the app sends what identifies it (the name you typed or confirmed, the product link you pasted or its product page, or the text read from your photo of the label) to our backend for the eligibility look-up, with the same kind of per-install token. If you check without an account, the app registers that token the first time you use it; it identifies the installation, not you. Items are read to answer you and are not kept.
Who else sees the metadata
Our backend forwards the product metadata to Google's Gemini API for a second classification opinion. Google receives the same metadata fields described above; it does not receive your user identity, your install's authentication token, or any other request context. Google's API processing is governed by their own privacy policy.
We do not sell or rent your data. We share it only with the providers named in this policy and with AI assistants you choose to connect (see below). We don't run advertising or tracking pixels. The operational telemetry and crash reporting we do run is described in the Telemetry section below.
Account sign-in and sync (optional)
Signing in is optional. The chip works fully anonymously, and eligibility decisions never require an account. If you choose to sign in (from the extension's popup, the "Sign in to sync" panel on the chip, or the mobile app), you can sync the eligible items you find across your devices.
We use Clerk as our authentication provider. When you sign in you provide your email address; Clerk emails you a one-time code and manages the login session. Clerk receives your email and standard authentication metadata (such as the verification attempt and session tokens). We never set a password, and we don't see or store your code.
Once you're signed in, the items the chip flagged sync to your Nodal account. What syncs is the same de-identified eligibility record described above (product metadata plus the verdict), tied to your account; never your browsing history, page contents, cookies, or payment information. Requests made while signed in carry a Clerk session token that identifies your account to our backend; signed-out requests carry only your install's revocable, non-identifying token.
Separately, if you ask to be notified about the Letter of Medical Necessity (LMN) workflow, you give us your email and, optionally, the item you were considering. We use it only to email you when that workflow is ready and to save the verdict you saw. It is never used for marketing, and you can ask us to delete it at any time.
AI assistants you connect (optional)
You can connect Nodal to an AI assistant, such as ChatGPT, Claude, or Meta’s Muse, so you can ask it whether something may qualify for HSA or FSA funds and keep track of what you checked. Nothing is shared with an assistant unless you connect it yourself and approve it on Nodal’s consent screen, where you also choose full access or read-only.
A connected assistant can:
- check whether products may qualify for HSA or FSA funds;
- see the items you’ve checked or saved in Nodal on any device: the product name, the store, and Nodal’s guidance;
- with full access only, save or dismiss items for you.
A connected assistant never sees your receipts, claims, letters of medical necessity, the amounts you paid, who an item is for, or your answers to health questions. Those stay in the Nodal app.
When an assistant asks Nodal to check a product, it sends product details: a product link, a barcode number, or a product name. We ask assistants not to send health information, we strip anything that looks like it before checking, and an unsaved check is kept for only 24 hours, so the assistant can explain or save it. When you save an item, we keep the product details (the cleaned product name, store, and any link path or barcode), never the rest of what the assistant typed. What Nodal returns goes to the company that runs the assistant (for example OpenAI, Anthropic, or Meta) and is covered by that company’s own terms and privacy policy, not ours.
For each connection we keep a record of which assistant it is, the access you chose, when you agreed, and when it last used Nodal, plus a short activity log of which actions it took and when (not their content). You can disconnect an assistant at any time at api.nodalhealth.ai/oauth/connections (sign in with your Nodal account), or by deleting your data in the Nodal app; it loses access within a minute.
Telemetry
The extension and beta app send operational telemetry events so we can find regressions: chip renders, unavailable on-device model counts, verdict buckets, merchant identifiers, the name of the product that was checked, capture-method categories, and latency measurements. Telemetry does not include full URLs, query strings, cookies, receipt bodies, account identifiers, or payment information.
By default, telemetry lives in a capped in-memory buffer on the backend and disappears when the server instance restarts. When persistent telemetry storage is enabled, events are mirrored into a capped recent-event log for the internal dashboard. The log is used for reliability and product-quality analysis, not advertising.
Crash and error reporting
The mobile app uses Sentry for crash reporting, configured to send as little as possible: no user identifiers (Sentry's default personal-data collection is switched off), no screenshots, no session replay, and breadcrumbs limited to an allowlist of app-lifecycle and navigation events. Crash reports carry the error message and stack trace, which is what makes a crash debuggable. Our backend separately sends error reports (error name, truncated message, stack frames, and coarse route tags; never request headers, bodies, query strings, or user fields) to a Sentry-compatible error-tracking service.
Launch and performance measurements
The mobile app sends launch and performance measurements (cold-start counts, app version, time-to-render and time-to-interactive timings, and platform metadata) to Expo's EAS telemetry services (expo-insights and expo-observe). These measurements carry no user content, no merchant data, and no health information.
Receipt capture
Nodal can ingest receipts to help substantiate HSA, FSA, HRA, or ICHRA-eligible expenses. Receipt capture is opt-in. Availability depends on the beta or pilot you join:
- Extension eligibility check. Available on product pages as you browse the web. This uses product metadata from the page; it does not read receipt contents.
- Email forwarding alias. Available in selected beta/pilot workflows. You forward receipt emails to an address we provision for you. We see only the emails you forward.
- Manual, photo, or share upload. Available in selected beta/pilot workflows. We process the receipt image, PDF, or text you choose to provide, plus any context you type into the app.
- On-device receipt reading. When you photograph a receipt in the mobile app, the image is read on your device by default, using Apple's on-device text recognition (Vision). The photo leaves your device only if the on-device read cannot produce a usable result, in which case it is sent to our backend to be read by Google's Gemini API. Either way, the read only pre-fills a draft: you review and confirm every value before anything is submitted.
- Gmail or Outlook connection. Deferred or pilot-enabled only. If enabled, you authorize Nodal to read a narrow set of receipt emails directly from your inbox via Google, Microsoft, or an approved email provider. We search only for messages matching a healthcare-merchant or receipt-pattern filter. We do not intentionally read unrelated inbox content.
Google user data: what we do, what we don't
Nodal's use and transfer of information received from Google APIs to any other app will adhere to the Google API Services User Data Policy, including the Limited Use requirements.
Specifically, if Gmail is enabled for your beta or pilot and you connect it to Nodal:
-
We use the
gmail.readonlyscope to read messages that match our server-side receipt filter, such as sender domain in a healthcare-merchant allowlist and subject lines containing "receipt", "order confirmation", "invoice", or similar terms. - The matched message body is sent to our LLM parser (Google's Gemini API) solely to extract structured receipt fields (merchant, total, line items, service date). Gemini's role is to provide the receipt-parsing feature you signed up for; no other use.
- We do not use Gmail data for advertising, sell it to third parties, use it to train any general-purpose AI model, or share it with anyone outside the narrow vendor list above.
- No human at Nodal reads your email content except in specific circumstances: with your explicit prior written consent when investigating a bug you reported, when required by law (and only the minimum necessary to comply), or for security investigations into abuse of our service.
- You can disconnect Gmail at any time from the Nodal settings page when that pilot workflow is enabled. Disconnecting revokes our OAuth token at Google.
Retention
- Structured receipt records (the parsed merchant + total + line items): 7 years, to match the IRS substantiation retention horizon for HSA/FSA expenses.
- Raw email bytes (the original message before parsing): 30 days, retained solely to recover from transient parser failures. Auto-deleted after 30 days.
- Extension and app telemetry: held in a capped recent-event buffer. When persistent telemetry is configured, the store is still capped to recent events rather than kept as a permanent behavioral history.
- Website inquiry records: kept while we manage beta and pilot relationships, and deleted on request unless a legal or security obligation requires us to keep them.
- Synced eligibility records (the de-identified product metadata plus the verdict, tied to your account once you sign in): 365 days from the point we receive them, after which they are deleted automatically. We also keep a separate record of each decision for product analytics, holding the verdict and how it was reached, the product name and merchant involved, and any reimbursed amount, with no account details and not tied to you. Those are kept on the same 365-day schedule.
- Your item list (one entry per product you checked or saved across your devices or a connected assistant: product name, store, and Nodal’s guidance): 365 days from the last change, or 30 days once you dismiss an item.
- Connected-assistant records: each connection is kept until you disconnect it, and ends automatically after 180 days without use. The per-connection activity log is kept for 90 days.
- On user-initiated deletion: receipt records we control are deleted within 24 hours where technically feasible, and OAuth tokens are revoked at Google for connected Gmail workflows. Deleting your data in the mobile app removes your synced eligibility records and your item list immediately, disconnects every connected assistant, and clears everything stored on the phone itself. That does not close your Nodal account, and it does not reach the copy the browser extension keeps in your own browser.
What we don't do
-
We don't run the extension on private or local pages (the
file://,chrome://, and Chrome Web Store origins are excluded). - We don't modify checkout flows. The chip is advisory only: it recommends, and you pay however you intended.
- We don't act as a healthcare provider. We don't issue prescriptions or letters of medical necessity. We don't see the items on receipts unless you give us a receipt in the app or through a beta or pilot workflow.
Your choices
You can uninstall the extension at any time from chrome://extensions. That
immediately stops all metadata transmission. You can also disable the on-device model in
chrome://flags if you'd like to use only the rules-based tier; the extension will
fall back gracefully.
In the mobile app, Account → Privacy and data has two controls: export a copy of your records, and delete them. Delete removes every check, claim, and setting stored on the phone, the eligibility records and item list synced to your Nodal account, and every connected assistant. It does not close your account, and the extension keeps its own copy in your browser until you uninstall it or clear its storage.
To see every AI assistant you’ve connected, what access it has, and when it last used Nodal, sign in at api.nodalhealth.ai/oauth/connections. Each one has a Disconnect button.
Children
Nodal is not directed at children under 13. We don't knowingly collect any information from anyone under that age.
Contact
Questions? hello@nodalhealth.ai.
Changes to this policy
We'll update the "Last updated" date at the top of this page when we change anything. If a change materially affects what we collect or how we use it, we'll surface a notice in-extension before the change takes effect.