Product preview · commercial release in preparation

Alsehli AIby Bahraini.online

Your AI.
On your machine.

Pay once. Use your installed AI for life. Private chat, images and voice on your own PC—with no recurring AI subscription and offline use after setup.

✓ Lifetime use✓ No AI subscription✓ Private on your PC✓ Guided setup

Watch the walkthrough Ask about early access

55.9 tokens/sMeasured generation average

16KLargest tested context

LocalChat, images and voice

Figures from an RTX 5060 Ti 16 GB test PC. Performance varies with hardware and workload.

The original desktop interface with chat, GPU monitoring and Settings — Inference stays local after the required assets are installed

Features Video Performance Requirements FAQ

CHAT / VISION / VOICE

One interface. More ways to work.

Features tested in the local application, with actions kept under the user’s control.

Chat on your own machine

Streaming Arabic and English conversations through a local Q6_K model, without a cloud AI provider.

Images and screenshots

Upload an image for a description or text extraction. Image description, OCR and screenshot understanding passed functional tests.

Speak and listen

Tap the microphone, speak and tap stop: your recognized message sends automatically. Local Arabic/English recognition and spoken replies, with no cloud speech provider.

Windows commands, with your approval

Open Notepad, Calculator, Explorer, Settings, Documents or Desktop. A separate confirmation is required for every action.

Make it feel like yours

Light and dark modes, a separate background color for each, and 12–24 px text. All appearance controls live in Settings.

See performance as it happens

GPU memory, activity, temperature and generation speed, with 4096, 8192 and 16384-token context settings.

PRODUCT WALKTHROUGH

See what you will use.

A 36-second visual walkthrough of the original interface, appearance settings and performance monitoring.

Chat and images
Voice and confirmed Windows commands
Light/dark mode and text size
GPU monitoring and generation speed

A captioned walkthrough made from actual app screenshots, not a live recording. Voice and command features are described rather than executed in the video.

36 seconds · silent, captioned product walkthrough

MEASURED, NOT ESTIMATED

Why we selected Q6_K.

Faster generation and more memory headroom on the test PC, with text and vision checks passing. Answer-quality equivalence between quantizations was not evaluated.

llama-bench results

3-repeat average

Tokens per second · higher is faster

015304560

Q6_K

55.91 tokens/s

Q8_0

44.65 tokens/s

512 prompt tokens / 128 generated tokens. Generation sample standard deviation: Q6 0.072, Q8 0.076 tokens/s; not a confidence interval.

GPU memory snapshot

Choose context size

GiB · total device usage, not model-only

0481216

Q6_K

13.42 GiB

Q8_0

15.49 GiB

Includes other applications running at measurement time. One snapshot per context setting, not a continuously measured peak.

Selected-context results: 4,096Actual prompt tokens: 3,513Q6_K: 53.29 tokens/sQ6 first token: 1.57 s

Test hardware, methodology and data

Measured 12 October 2026: RTX 5060 Ti 16 GB, i5-10400 and 64 GiB RAM on Windows 11. llama.cpp b11429 with CUDA 13.4, 6 threads, batch 512, microbatch 256, Flash Attention and Q8 KV cache.

The speed benchmark uses 3 repeats with the server stopped. Context tests use uncached long prompts and 128 generated tokens. First-token time includes long-prompt processing. 16384 tokens passed a functional test, not a prolonged soak test.

Download benchmark data (JSON)

Light-mode Settings with background color and font-size controls

PERSONAL BY DESIGN

Make it comfortable to read.

Choose light or dark mode, change the background and adjust text size. Preferences are saved locally without changing the model configuration.

All appearance options inside Settings
A separate background for each mode
12–24 px font size
One-click appearance reset

BEFORE YOU INSTALL

Is your PC ready?

The machine we tested, alongside clear starting recommendations. Minimum specifications and other hardware have not yet been validated.

Operating system

Windows 11 · x64

Windows 11 Pro is validated. Windows 10, macOS and Linux are not tested in this release.

Graphics

NVIDIA · 16 GB VRAM

RTX 5060 Ti 16 GB with CUDA was tested. 16 GB is our starting recommendation; other GPUs and CPU-only use remain untested.

Memory and storage

32 GB RAM recommended

The tested PC has 64 GiB RAM. Plan for around 20 GB free storage; Q6 plus the projector alone are 8.74 GiB. This is planning headroom, not a validated minimum.

Voice and setup

Microphone and speakers

Windows microphone permission is required. Internet is needed for initial assets; normal inference runs locally afterward.

The API stays on 127.0.0.1. The model does not receive unrestricted OS access. The current build reuses a speech environment on the test PC; a standalone installer is being prepared.

THE DETAILS

Before you get started.

Is this an official DeepSeek release?

No. The application is by Bahraini.online and uses Jackrong’s community Qwen3.5-9B fine-tune. The underlying model name does not make this an official DeepSeek product.

Does my inference data go to a cloud AI service?

Chat, images, speech recognition and synthesis run locally after installation. Website and email links open when you choose them. This product page does not connect to your PC.

What can the Windows commands do?

Only six allowlisted open actions, each with a Yes/No confirmation. No arbitrary shell execution or file deletion. The confirmation policy was tested automatically; real execution waits for the user’s approval.

How was speech tested?

English and Arabic synthesized speech was transcribed successfully. Real microphone open/capture/close was checked. Accuracy across accents and noisy rooms has not been comprehensively evaluated.

Can I download or buy it now?

This is a product preview. We are preparing standalone installation, license notices and an open-source release with paid downloads or support. Installation and setup are available for $19 USD after hardware compatibility is confirmed. The standalone downloadable installer is still being prepared.

SETUP FOR ONE PC

Alsehli AI installation & setup

$19 USD · One-time payment · lifetime use of the installed version

Private AI without a subscription: chat, images and voice are processed on your PC. After downloads and setup, normal inference works offline. Payments, downloads and contacting support need internet.

Includes local AI and NVIDIA GPU configuration, text and image testing, Arabic/English voice setup, and 7 days of installation-related support. Lifetime use applies to the installed version; it does not include unlimited support or a promise of all future upgrades.

Contact us through the support form or hello@bahraini.online to confirm PC compatibility and arrange installation before paying. This is an installation service for one Windows 11 PC, rather than an instant app download.

Pay with PayPal — $19 USD

Secure checkout opens on PayPal. This website does not collect your card details.

Check PC compatibility first

HERE TO HELP

Contact support.

For installation, questions or help with the app. Please do not send passwords or payment details.

hello@bahraini.online

BAHRAINI.ONLINE

Local intelligence. A clearer experience.

Interested in early access, installation or support? Talk to us.