Press kit

What Fihris is, who makes it, what it costs, and how its AI works — in the short form, with assets to download.

Last updated 17 September 2026

On this page

In one paragraph #

Fihris is a research workspace for iPhone and iPad. You find papers across the major scholarly sources, read and annotate them with Apple Pencil or a fingertip, ask questions about a whole collection and get answers that point at the exact paper and page they came from, and write the manuscript in LaTeX citing the same library. It replaces the usual stack of a reference manager, a PDF annotator and an online LaTeX editor with one app — and adds a grounded AI layer that none of them has, without ever putting your papers on somebody else's server.

In one line #

Fihris — your research library, read, understood and cited, on your iPad.

Facts #

NameFihris (فهرس), pronounced FIH-ris
CategoryReference manager · PDF reader and annotator · LaTeX authoring · research AI
PlatformsiPhone and iPad, iOS and iPadOS 18 or later. One universal app — the phone build is the full app, not a companion
Mac / AndroidA Mac version is intended, unannounced. No Android version is planned
DeveloperFihris LLC, Saudi Arabia
PriceFree to download and use, including all AI on your own device, Apple's private cloud or your own API key
Fihris ProOne-time purchase, $22.99. Unlocks iCloud sync, mind maps, analysis, manuscripts and Export everything
Fihris IntelligenceOptional subscription, $9.99/month or $89.99/year, 7-day free trial, 300 credits a month. Other storefronts priced by Apple
Availability[APP STORE URL] · released [LAUNCH DATE] · current version [CURRENT VERSION]. Not currently offered in the 27 EU storefronts
Press contact[email protected]

What it does #

  • Discover — one search across arXiv, OpenAlex, Crossref, Semantic Scholar, DBLP, CORE and Unpaywall, deduplicated, with open-access PDFs pulled straight to the device.
  • Read — a PDF reader with highlights, underlines, strikethrough, Apple Pencil ink, a reflowed reading mode for small screens, OCR for scanned papers, and notes linked to the exact span they came from.
  • Ask — questions about one paper or a whole collection, answered from passages retrieved out of your own library, with a citation chip on every claim that opens the paper at that page with the passage highlighted.
  • Mind maps and analysis — a hand-editable map of your own ideas, and a factual view of who cites whom among papers you choose.
  • Write — a LaTeX editor with a cite picker over your library, bibliography export in BibTeX, RIS and CSL-JSON, and a one-tap hand-off to Overleaf.

The four routes an answer can take #

Fihris always uses the smallest route that fits, and shows you which before it sends anything.

  1. On this device — Apple's on-device model. Nothing leaves the device. Free, works offline.
  2. Apple's private cloud — Apple's Private Cloud Compute, inside Apple's privacy boundary. Free.
  3. Fihris Intelligence — a relay we operate that stores no question, passage or answer, forwarding to Anthropic's Claude. Paid, optional, and gated behind a consent card that names both the relay and Anthropic.
  4. Your own Claude or ChatGPT account — if you connect a key, straight from device to provider. Your account pays. Free to Fihris and to us.

The privacy position, stated plainly #

  • Papers never reach a server we run. Fihris's relay receives short retrieved passages, never a PDF and never a whole paper, and the relay keeps nothing — no question, passage or answer is stored there, in any environment. What it does keep is a credit ledger with no content in it, and one technical line per request in Cloudflare's logs — task type, model, token counts, timing, status — again with no content. The relay's "keeps nothing" is a statement about the relay, not about Anthropic: Anthropic's default API retention is 7 days and Fihris does not claim otherwise.
  • The search index and the text embeddings are built on the device and never leave it, on any route, on any plan.
  • There is no account and no sign-in. No email address, no profile, no password.
  • There are no analytics. No analytics SDK, no advertising SDK, no crash reporter, no tracking.
  • Sync is the user's own iCloud, off by default, in a private database the developer cannot read.
  • Every third-party route asks first, with a card that names the processor and lists exactly what will be sent.
  • An on-device log, Sent to your account, records every send that left the device — route, paper titles, page and token counts, never content — for the user to audit.

Why citations can't lie about the page #

Fihris chunks every PDF itself, so it owns the mapping from a passage to its paper, page and character range. Models are handed numbered passages and are asked to cite by number; they are never asked for, and never permitted to write, a page number. Fihris resolves the numbers to pages from its own index. A number a model invents resolves to nothing and never becomes a citation.

The consequence is the interesting one for a research tool: the page pointer cannot be hallucinated. Only the choice of passage can be wrong, and that is one tap away from being checked.

The name #

Fihris (فهرس) is Arabic for index or catalogue. It nods to Ibn al-Nadim al-Warrāq's Kitab al-Fihrist of 987 CE, a catalogue of some 4,300 authors and arguably the first reference manager ever compiled. Please use it untranslated; it is not an acronym.

Assets #

  • App icon (1024 px PNG): [APP ICON ASSET]
  • Screenshots, iPad and iPhone, light and dark: [SCREENSHOT SET]
  • Logo and wordmark: [LOGO ASSET]

All assets may be used in coverage of Fihris without asking. Please do not alter the icon or the wordmark.

Contact #

Press and everything else: [email protected]