Privacy policy
What Fihris does with your data. Almost all of it never leaves your device, and this page says exactly where the rest goes.
On this page
Fihris is a reference manager, PDF reader and LaTeX authoring app for iPhone and iPad. Most of what you put into it never leaves your device. This page says exactly what does, when, and to whom.
Who is responsible for your data #
Fihris is made and operated by Fihris LLC, Saudi Arabia. That is the data controller for everything described here.
Contact: [email protected]
What stays on your device, always #
- Your Library: papers, PDFs, metadata, tags and collections.
- Your highlights, underlines, strikethroughs, Apple Pencil ink, notes and manuscript drafts.
- The search index and the text embeddings Fihris builds over your Library. These are created on your device with Apple's Natural Language framework. They are never uploaded, on any plan, on any route.
- Questions answered by the model that runs on your device.
If you turn on iCloud sync, this same data is also copied into your own private iCloud account — a private CloudKit database and ubiquity container that belongs to you. We cannot read it. It is not a server we operate, and Apple's private database is inaccessible to app developers by design.
There is no account and no sign-in in Fihris. We do not know who you are.
We do not operate a server that stores or serves your papers. That is enforced by the architecture, not only by this page: the one server Fihris runs is described below, and it is built so that it cannot keep your papers even if it wanted to.
Where answers run, and what each route sends #
Fihris always tries your device first. Four routes exist, and every screen tells you which one it will use before it sends anything.
1. On this device. Apple's on-device model. Nothing leaves the device.
2. Apple's private cloud. Apple's Private Cloud Compute. Your question and the passages Fihris retrieved from your own papers are processed by Apple for that one request. Apple states that this data is not retained and is not available even to Apple's own staff. This route is disclosed in Settings ▸ AI; it is not a third-party route, so it does not ask for separate consent.
3. Fihris Intelligence (the subscription). Your question and the specific passages Fihris retrieved from your own Library are sent to Fihris's relay, a small server we operate. The relay:
- never receives a PDF or a whole paper — only the short excerpts our on-device search already selected as relevant to your question;
- never fetches anything from arXiv, a publisher or any other source on your behalf;
- keeps nothing — no question, excerpt or answer is stored at the relay, in any environment. It is never written to a database, a log file or a backup there;
- forwards your question and excerpts to Anthropic (the maker of Claude) to generate the answer, and streams the answer back to your device.
That "keeps nothing" is a statement about Fihris's relay only. It is not a claim about Anthropic — see the next paragraph for what happens at Anthropic's end, which is a different thing with a different answer.
What we hold about a relay request, and what Cloudflare holds #
Two different things, in two different places, with two different answers to "can you delete it?"
(a) Our ledger — Fihris holds it, and we can delete it on request. The relay keeps a small database that exists so the subscription can be metered and abuse stopped. It holds: an opaque install identifier, the App Attest key registered to that install, your subscription's App Store transaction identifier, the current billing period and how much of the allowance it has used, any credit top-ups and what is left of them, per-hour and per-day credit totals, daily counts of requests and of refusals, and what Apple last told us about your subscription's status. It holds no question, no excerpt, no answer, no paper title and no per-request record of what you asked. This is the data we can give you a copy of, correct, or delete — see "Your rights" below.
(b) Operational logs — Cloudflare holds them, and we cannot delete an individual line. To run the service at all, the relay writes one technical line per request into Cloudflare's own logging systems (Cloudflare Workers Logs, and Cloudflare AI Gateway's metadata logs). That line contains: the time, the opaque install identifier, the transaction identifier, the task type (for example "grounded question"), the model used, token counts, how long it took, whether it succeeded or was refused, and the true/false results of the automated scope check. It never contains your question, your excerpts or the answer — the relay is written so that only that fixed list of fields can be emitted at all, and there is deliberately no setting anywhere that would log the content.
Those lines live in Cloudflare's systems, are retained under Cloudflare's own retention and storage limits rather than ours, and age out on Cloudflare's schedule. We do not control that schedule and we cannot promise to delete one specific line on request. If you ask us to delete your data, we delete your ledger rows (a), and the log lines (b) fall away with Cloudflare's normal retention. We would rather tell you this plainly than imply a power over someone else's log store that we do not have.
Anthropic #
Anthropic acts as our processor for that one request. Anthropic is not us, and Anthropic's own privacy policy and terms — not this page — govern what Anthropic itself keeps. Anthropic's default retention for API requests is 7 days. We have asked for a shorter, zero-retention term; until Anthropic confirms one in writing for this account, the 7-day default is what applies, and we will not claim otherwise here. Anthropic's terms also allow it to keep automated safety-classification results — not the content of your question — regardless of the retention term, so that it can enforce its own usage policies. Anthropic's current detail is at privacy.claude.com.
Cloudflare #
The relay runs on Cloudflare (Cloudflare Workers and Cloudflare AI Gateway). Cloudflare is our hosting and gateway provider and a processor for this traffic. The gateway is configured to store no request or response content — the question, the excerpts and the answer are not written to Cloudflare's logs. What Cloudflare does log and retain is the per-request metadata in (b) above, under its own retention terms.
4. Your own Claude or ChatGPT account. Only if you connect your own API key. Nothing is sent until you approve an on-screen consent card that names the provider and lists exactly what would be sent. This route goes straight from your device to that provider using your key; Fihris's relay is never involved, and your own account pays for it. Your key is stored in the iOS Keychain on your device only.
Every route past your device is your choice. You can always decline and get an answer from your device or from Apple's private cloud instead, or connect your own key.
Consent #
Before the first send on Fihris Intelligence or on your own key, Fihris shows a consent card naming the route, the processor, and exactly what will leave the device — your question, how many excerpts from how many papers covering how many pages, and, where applicable, paper titles, notes or highlights. Nothing is sent before you approve it. You can approve once, approve for a collection, switch to Apple's private cloud, or cancel. Consent choices can be reset in Settings ▸ AI.
If what Fihris sends ever widens, the consent card is rewritten and every existing approval is invalidated, so you are asked again.
The "Sent to your account" log #
Every time a question is answered by anything other than the model on your device, Fihris writes an entry to a log on your device: which route was used (Apple's private cloud, Fihris Intelligence, or your own account), the titles of the papers involved, how many pages and excerpts, and token counts. It never contains the question text or the answer text. You can read it, and clear it, in Settings ▸ AI ▸ Sent to your account. This log is never sent anywhere. It exists so you can audit what Fihris has done on your behalf.
Purchases #
Fihris is sold through Apple's App Store. Apple is the merchant. Apple processes the payment and gives Fihris a signed transaction record, which the relay uses to verify your entitlement and meter your credits.
We never see your name, your card number or your billing address. Apple does not give app developers that information. What we hold is the transaction identifier and your credit usage — the ledger described in (a) above. Apple's own privacy policy covers what Apple collects to process the purchase.
Device verification (App Attest) #
Requests to the relay carry an Apple App Attest assertion, which lets the relay confirm that a request came from a genuine, unmodified copy of Fihris on a real Apple device. It is a check on the app, not on you: it carries no name, no account and no identity, and it is used only to stop someone using the relay without paying for it.
Analytics #
There are none. Fihris contains no analytics SDK, no advertising SDK, no crash-reporting service and no tracking of any kind. Nothing about how you use the app is collected for marketing, profiling or measurement. The relay's ledger exists to bill for credits and to stop abuse, and nothing else reads it.
Who else sees your data, and why #
| Who | What they see | Why |
|---|---|---|
| Apple (App Store, StoreKit) | Your purchase and subscription status | Processing your payment |
| Apple (private cloud) | A question and retrieved excerpts, for that one request | The route you chose, when you choose it |
| Anthropic (through Fihris's relay) | A question and retrieved excerpts, for that one request | The route you chose, when you choose Fihris Intelligence — always after you consent |
| Anthropic or OpenAI (directly) | A question and retrieved excerpts, for that one request | Only if you connect your own key, always after you consent |
| Cloudflare | Nothing of the content. Cloudflare hosts the relay, and its own logs keep the per-request metadata in (b) above — timestamp, install identifier, transaction identifier, task type, model, token counts, latency, status — retained under Cloudflare's terms, not ours | Hosting and running the relay: operations, billing, rate limits, abuse prevention |
| arXiv, OpenAlex, Crossref, Semantic Scholar, DBLP, CORE, Unpaywall and the other sources Discover queries | Your search terms, when you search | Finding papers. arXiv's terms do not permit us to store or re-serve papers, so we don't — a PDF is downloaded straight to your device |
Anthropic and Cloudflare operate infrastructure outside Saudi Arabia. Your question and excerpts may therefore be processed on servers outside the Kingdom for the seconds it takes to answer, and nowhere else.
This transfer happens only when you choose Fihris Intelligence or connect your own key, and only for the seconds it takes to answer — the same performance-of-service basis (PDPL cross-border transfer regulation, Article 29) that lets any cloud-backed feature work at all. We have not yet put a data-transfer agreement (such as standard contractual clauses) in place with Anthropic or Cloudflare specific to Saudi Arabia; this is being reviewed with Saudi counsel before the app is submitted for review in the Kingdom.
Children #
Fihris is not directed at children and does not knowingly collect data from them beyond what this page already describes. There are no accounts, no profiles, no advertising, and nothing here is collected for marketing. If you believe a child has used Fihris Intelligence without appropriate permission, write to [email protected] and we will address it.
Your rights #
Saudi Arabia's Personal Data Protection Law (PDPL) gives you the right to know what is processed about you and why, to get a copy of it, to have it corrected, to have it deleted, and to restrict or object to processing. You can also complain to the Saudi Data & Artificial Intelligence Authority (SDAIA).
Because Fihris has no account, the only thing we hold that is tied to you is the ledger described in (a) above. Everything else — your Library, your notes, your papers — never left your device or your own iCloud, so there is nothing else for us to hand over, correct or delete.
To exercise any of these rights, email [email protected] with your App Store transaction identifier. You can find it in your purchase confirmation email from Apple, or in Settings ▸ AI ▸ Fihris Intelligence. We use that identifier to find your ledger rows; we have no other way to identify you, and we will not ask you to create an account in order to prove who you are.
What a deletion request does and does not reach. We delete the ledger rows in (a). We cannot individually delete the operational log lines in (b), because they sit in Cloudflare's own logging systems under Cloudflare's retention rather than ours; they age out on Cloudflare's schedule. Those lines contain no question, excerpt or answer. If you need a firmer answer than that for a formal request, say so in your email and we will go back to Cloudflare about it rather than guess here.
If you are elsewhere in the world, similar rights (for example under the GDPR, where it applies) can be exercised the same way. Fihris is not currently offered in the 27 EU storefronts; see the App Store listing for current availability.
Changes to this page #
If what Fihris Intelligence sends ever widens — a new task type, a new kind of paper content, a new processor — this page, the in-app consent card and the App Store privacy label all change together, before the change ships, and the consent card asks you again. The date at the top of this page is updated whenever it changes, and material changes are announced in the app's release notes.