An app by Umber
on-device · audio intelligence

Hujam.

Words you
can actually
own.

Hujam records, transcribes, translates and writes reports — entirely on your device. No cloud mandated. No leaks. The models live on the phone. The audio never leaves it unless you need it. Encrypted data at rest.

Local STTLocal MTLocal LLMNetwork · On Demand
9:41
5G·
BACKREC
ATHENS · EXARCHIA CAFÉ
EL → EN2 MICSFLAC · 48KON-DEVICE
00:03:09
Recording · 1.4 MB · 18.2 GB free
LIVE TRANSCRIPTON
Good. Then I can speak freely. The cases I described last week — there are now forty-three of them
00:01:27Topic shift
00:02:22Off the record
00 / The promise

Yours.
On Device!

On-device focus

0bytes

sent off your device, without your consent. We can't even build sync without your explicit per-file consent.

On-device focus

1accounts

Simple sign-up. No spam. No cloud. The app just works.

On-device focus

6models

Speech, translation and LLM — all downloaded once, run locally, swappable.

01 / How it works

Capture.
Understand.
Report.

01 / Capture

Hit record. Stay in the room.

Two-mic capture, FLAC at 48 kHz, with airplane-mode-friendly buffering. Topic shifts and off-the-record markers as you go.

02 / Understand

Transcribe and translate on-device.

Curated Hujam on-device Models for transcription & translation. Speaker diarization included. Nothing uploads — the work happens on the phone.

03 / Report

Hand it a brief. Get a draft.

A local LLM turns the transcript into the document you actually needed — interview brief, court note, field summary, captions.

02 / Features

All voice
workflows need!
Nothing more.

On-device by design

Audio, transcript, translation and report never have to go to the server. You decide when and how to export

T
Transcription Models

99 languages, mixed-language passes, accent-aware decoding. Light Whisper Small for older devices and short clips.

Translation Models

Bidirectional translation across 200 languages, integrated into the transcript view in real time.

Speaker diarization

Identify and label speakers across long recordings. Rename once — Hujam remembers.

Local LLM reports

Local LLM Models draft your interview briefs, hearing notes, summaries and captions, on the phone.

No-cloud architecture

Data is on device & encrypted. You decide when to rely on cloud providers.

03 / Model library

Run them.
On here.

Storage · 12.4 GB of 64 GB
Models 6.7 GBRecordings 2.1 GBCache 2.0 GB
Speech
Default
Whisper Large v3
Size
1.5 GB
Coverage
99 languages
Speed
slow
Speech
Light
Whisper Small
Size
244 MB
Coverage
99 languages
Speed
fast
Speech
Streaming
Parakeet TDT 1.1B
Size
600 MB
Coverage
EN, ES, FR, DE
Speed
very fast
Translate
Default
NLLB-200
Size
2.4 GB
Coverage
200 languages
Speed
medium
Reports
Default
Qwen2.5 7B
Size
4.7 GB
Coverage
Multilingual
Speed
medium
Reports
Light
Phi-3.5 Mini
Size
2.2 GB
Coverage
EN, ES, FR, DE
Speed
fast
04 / Built for

For people
whose sources
trust them.

Journalists
Interview a source. Hand them the phone if they want to see "Network · off". File the brief on the train home.
Lawyers
Bilingual client intake — Greek, Arabic, Tigrinya — without sending a single packet to a third-party transcription service.
Field researchers
Record interviews in Lampedusa, Athens, Beirut. Process and translate on the device, even offline.
Newsrooms
Hearing transcripts, embargoed press briefings, anonymous sources — kept off shared drives by default.
Classrooms
Hearing lectures, class notes, briefings, action items — kept on device or exported on demand
05 / Ships on

Mobile first.
Then everywhere.

A
Android — Play Store · Available
iOS
iPhone · iPad — beta soon
macOS — Q4 2026
Windows — Q4 2026
W
Web — by design, never
Early access

Sideload the
private beta.

We're letting in new cohorts every two weeks. Tell us a bit about how you'd use Hujam and we'll send an APK and an invite to the next cohort.

Or just say hi

hujam@
umberab.com

Open contact form