Mufkerع

Mufker

Press Enter to send · the circles open a tool and run nothing

What Mufker is

An Arabic workspace — text, voice, documents and images.
It reads what runs long, delivers what ships, and shows where it came from.

General tools write a good answer. Mufker writes one and opens what is underneath: which page was read, where this sentence sits in your file, and what was never read at all.

Because what you put your name under needs more than a good answer.

It tells you what it read — and what it did not

Every task comes with a coverage map: how many pages were processed out of how many, and which parts were skipped and why. Citations open the exact place in your file — and any that cannot be opened are removed and counted, because a citation you cannot open is worse than none.

The whole book at once

Up to 4,000,000 characters in one request and 5,000 pages per document. That alone is what makes it possible to unify a term across a book or catch a contradiction between chapter one and chapter nine — which a chat reading one chapter at a time never sees.

Arabic from the roots, not translated

Dialect stays as it was spoken instead of being turned into MSA. Counted nouns agree correctly. Search sees past hamza, diacritics and tatweel, so «اطار» finds «إطارًا». This is not an English interface that was translated.

An Arabic voice that reads, not spells

Thirty voices with accents and tones, effects performed where you place them — a laugh, a sigh, a pause — and one performance across long takes, every part checked before it is joined. Long-form reading up to 600,000 characters, with a report of how many parts were joined and where the seams fell.

Code decides; the model judges

A terminology change is proposed by the model and applied by code, so an unrequested edit is impossible, not merely unlikely. Diacritised text is stripped and compared with the original letter by letter. Table and invoice totals are re-added from their cells. What can be decided is decided, not left to judgement.

Many models, agents working together

Each task has its own route, model and fallback — chosen as data, not code. Heavy work is spread across parallel agents: a thirty-thousand-character book chapter comes out as one file in seventeen seconds.

Search that shows its sources

Web search is a tool called when needed, not always, and sources are shown as they came back. Their absence is information too: when it did not search, it says so instead of leaving the impression.

Pay for work, not for a seat

A task that fails is not charged, the held estimate returns in full, and nothing beyond the hold is ever taken — an overrun is our error, not your debt. Your balance stays yours; it does not expire at the start of the month.

Arabic audio studio

Your text in a real Arabic voice, not a machine spelling it out

Six voices, each reading its own line. Press one.

Accents, and dialect when you want it

Egyptian, Levantine, Gulf and Maghrebi accents on your words as written — and “Convert to dialect” rewrites the script in dialect for you to review before it is spoken.

Performance directed in words

Scene, context and performance notes in plain language, plus effects where you place them: a laugh, a sigh, a pause. Direction is never read aloud.

One voice, however long

Long takes are read in parts on the same engine, seed and brief; each part is checked before it is joined, and levels are matched.

A whole chapter, one file

Split at sentence ends, joined without re-encoding, with a report of how many parts it was assembled from.

See all audio tools+5

Arabic Text Tools

Long text, and tools that know what to do with it

Proofreading, rewriting and summaries — plus what an editor reading one chapter at a time cannot do: unify a term across a book, catch a contradiction between two chapters, and add diacritics with proof that no letter changed.

See all of Arabic Text Tools+3

Arabic Document Tools

Your files are read — and you are told which page was

PDF, Word, Excel and PowerPoint as text and as tables with checked totals; a summary where every point carries its page; and a comparison of two versions computed from the characters, not from an estimated reading.

Arabic Image Tools

An image is read, with what was read kept apart from what was inferred

The image's text as it is; a chart explained with what was printed kept apart from what was estimated; alt text in Arabic and English; invoices with the arithmetic redone; handwriting with uncertain words highlighted.

See all of Arabic Image Tools+1

Comparison

Not a replacement for a general model — what gets built on top of one

Mufker calls the same models you already use. The difference is not who answers but what remains after the answer: evidence you can open, arithmetic that is checked, and Arabic that was never translated.

MufkerChatGPTClaudeGeminiDeepSeek

How many pages were processed out of how many, and which parts were dropped and why. In a general chat you cannot tell whether it read your whole document or its beginning.

— means “partly”, not “no”. Services were compared on what their interfaces advertise in 2026; the last column groups all four because they converge on these rows.

Plans

Pay for work, not for a seat

Your credit balance is charged for what a task actually consumes — a page of text costs a fraction of an audiobook chapter. A task that fails is not charged at all.

Every account refills weekly — not monthly — so a generous week is not followed by three lean ones

Free

With a verified phone number

Free

6500credits a month

Covers any one of these a month:

270 chat conversations

11 hours of transcription

2 audiobook chapters

  • 1,500 credits every week
  • All four sections, in full
  • 30 pages per document
  • 30,000 characters of narration
Start free

With a Google or GitHub sign-in — no phone number

Mufker Plus

For regular work

$5/ month

21500credits a month

Covers any one of these a month:

895 chat conversations

37 hours of transcription

7 audiobook chapters

  • 15,000 credits a month, on top of the weekly refill
  • 300 pages per document
  • 120,000 characters of narration
  • Three tasks at once

15000 on the monthly renewal, plus 1500 every week.

Coming soon

Final price; payments open soon

Best for everyday work

Mufker Pro

For translators, editors and consultants

$15/ month

74000credits a month

3.4× what Plus gives

Covers any one of these a month:

3083 chat conversations

128 hours of transcription

26 audiobook chapters

  • 67,500 credits a month, on top of the weekly refill
  • 1,500 pages per document
  • 300,000 characters of narration
  • Six tasks at once
  • DOCX, PDF and SRT export

67500 on the monthly renewal, plus 1500 every week.

Coming soon

Final price; payments open soon

Mufker Studio

For teams and heavy audio users

$49/ month

226500credits a month

10.5× what Plus gives

Covers any one of these a month:

9437 chat conversations

394 hours of transcription

82 audiobook chapters

  • 220,000 credits a month, on top of the weekly refill
  • 5,000 pages per document
  • 600,000 characters of narration
  • Twelve tasks at once
  • 200 GB storage and queue priority

220000 on the monthly renewal, plus 1500 every week.

Coming soon

Final price; payments open soon

No account needed: 300 credits to try it · All the details and questions

Frequently asked

What people ask before they hand over a long file

What exactly is a “credit”?
Mufker's unit of account. A task deducts what it actually consumed at the provider, not a flat price per request: a page of text costs a fraction of what an audiobook chapter does. What you see deducted is what was really used, with no intermediate unit of our own invention.
Why is there no seven-day free trial?
Because the free plan is better. It renews every week, with no card and no end date, and uses the same models. A timed trial pushes you to test the tool in a busy week and then ends on the day you needed it — a race against a clock, not a test of the product.
What happens when my balance runs out?
New tasks stop; nothing is deleted and no running work is cut off. The weekly allowance returns on schedule, and a task that fails — for any reason — is not charged at all: settlement never exceeds what was held, and an overrun on the estimate is our error, not your debt.
What if the connection drops during a long transcription?
Nothing is lost. The recording is split into windows and each is saved on the server as soon as it completes, so the task resumes from the last completed window and you are not charged twice. If a window fails, it appears in the transcript as a declared gap — never silently dropped.
How do I know an answer is really based on my file?
Every sentence grounded in a source carries a citation that opens the exact place — a moment in the audio or a page in the document. A citation that does not resolve is removed and listed in the result's limitations, because a pointer to nothing is worse than none. Every result has a receipt naming the model that ran, the sources read and what could not be read.
Does it read the whole document or just the beginning?
It reads what fits the model's window and tells you which pages were read and which were not. Anything left out for space appears in the coverage map by name and page, not as a vague note. That — not polished prose — is the problem Mufker was built around.
Do you keep my data or train on it?
No training on your data. Files are stored privately and only reachable through short-lived signed links. You can export or delete at any time, and for sensitive work you can choose a restricted privacy route that keeps it away from providers outside a defined list.
Does it handle Arabic dialects?
Yes, and it keeps them rather than converting to Modern Standard Arabic — Gulf, Levantine, Egyptian and Maghrebi. Where speech is unclear the segment is marked low-confidence instead of passed off as certain, and if the engine reports no confidence, none is invented.
Can I download the result and work on it in Word?
Yes: DOCX, PDF, TXT, and SRT/VTT for transcripts, with correct right-to-left formatting rather than reversed text, and a choice of Arabic font for the file.
How is it different from a general chat like ChatGPT?
For general work we claim no advantage — they are better there, and Mufker calls the same models. The difference shows on long Arabic material that has to prove its source: a coverage map of what was read, citations that open the place, a receipt for every result, and tools that apply by computation rather than suggestion, like unifying a term across a book or comparing two versions character by character.
When do payments open, and how are prices calculated now?
The prices shown are final, and payments open soon once the payment provider's checks complete; until then paid plans show “coming soon” and the free plan works in full. The “how far it goes” figures on each card are computed from today's provider prices, not a marketing estimate, so if a price changes the figure changes with it.
Why is a phone number needed for the weekly renewal?
Because whatever renews for free is what duplicate accounts target, and transcription and generation have a real cost paid to a provider. You confirm once, and the allowance then returns weekly on its own. You can use Mufker without an account first and decide afterwards.

Question not here? Write to us — we read every message.