Students
Start on the Free plan. Ask in Uzbek or Russian and get explanations, summaries and exam prep in your own language.
All in one place
1750+ models. From every AI family.
Claude, GPT, Gemini, DeepSeek, Qwen, Kimi and more — all in one picker. Switch models instantly and the conversation carries on.
Loading…
New models appear here the day they launch.
How it works
From sign-up to your first answer in a minute.
- Step 1
Create a free account
Sign up with email or Google. No card needed — the Free plan includes Llama 3.3, Gemini 2.0 Flash and DeepSeek R1.
- Step 2
Pick a model — or let Auto choose
Switch between Claude, GPT, Gemini and others mid-conversation. Context and memory carry over.
- Step 3
Work on web, desktop or CLI
Use one account in the browser, the desktop app and the terminal code agent. Upgrade only when you need more.
New
Your workspace, on your terms
Crownwork 0.9 and the latest on the web.
- Crownwork 0.9
Crownwork Mods
Make Crownwork yours: describe it in plain words and AI builds a theme, a quick command or your own panel (timer, kanban…). Mods run in a sandbox with no access to your files, the internet or your account. The best ideas land in a gallery for everyone.
- Crownwork · Ollama
Sovereign Mode — offline agents
Local agents in Crownwork run on a schedule with an Ollama model on your computer. Requests only go to 127.0.0.1 and results stay on your machine — even with Wi-Fi off. Requires Ollama and an installed model.
- Web · Telegram
Agents + Telegram
Personal agents on a schedule: daily news, official Central Bank rates, exam questions, Gmail summaries. Results arrive on the site and in Telegram; reply to an agent's message to keep going.
- UzBench v1
UzBench — open leaderboard
We openly measure how well AI models handle Uzbek: 45 questions, automatic grading, no judge model. Questions, answers and method are all public.
See the leaderboard
Download
SOVEREIGN on your computer
Crownwork, the desktop app for everyday work, or the sov CLI right in your terminal. Free for Windows, macOS and Linux.
Windows
Windows 10 / 11 · 64-bit installer
macOS
macOS 12+ · disk image
Not sure? Apple menu → About This Mac → Chip.
Linux
x86_64 · portable, no install
Builds aren't code-signed yet — on first launch:
- Windows SmartScreen: click “More info” → “Run anyway”.
- macOS: try to open the app once, then go to System Settings → Privacy & Security → “Open Anyway”.
- Linux: chmod +x the .AppImage, then run it.
Windows
PowerShell · no admin rights needed
or download sov.exeirm https://soveregn.xyz/install.ps1 | iexmacOS / Linux
Terminal · bash / zsh
curl -fsSL https://soveregn.xyz/install.sh | shor download the binary:
npm
Any OS · needs Node.js 20+
npm install -g @islombekrrr/sov-cli
Then run:
sovWhy SOVEREIGN
Privacy, memory and sources in one account.
- Blind Prompting
Names masked before sending
Turn on Blind Prompting and names, numbers and companies in your messages are masked in your browser before the request goes out.
- Memory
Memory you can edit
It remembers what you tell it. You can view, export or delete any of it.
- Verified
Verified answers
Search answers come with real source links. A second AI checks key claims against those sources and flags anything it can't confirm.
- Auto
Auto model choice
Auto picks a model for each request, balancing quality and price.
Who it's for
Made for Uzbekistan and the CIS.
The same models used worldwide, with Uzbek and Russian interfaces. Pay by Visa/Mastercard, crypto or SBP.
Developers
A terminal code agent, a desktop app and 1750+ models behind one account — compare models on the same task in seconds.
Businesses
Blind Prompting (when on) masks client names, numbers and companies before a request leaves SOVEREIGN; chats and memory can be deleted at any time.
Interface languages
- O'zbekcha
- Ўзбекча
- Русский
- English
Payment methods
Card — via Dodo Payments: Visa, Mastercard, American Express, Discover, JCB, UnionPay, Apple Pay and Google Pay; monthly/yearly subscription. Humo and Uzcard aren't supported yet — use a Visa/Mastercard enabled for international payments. Crypto and SBP — via RollyPay, one-time payment.
Models
Switch models mid-conversation. Context carries over.
A current flagship from each major lab. The model picker has 1750+ more.
- Plan: Ultra
Claude Opus 5
Anthropic
- Plan: Ultra
GPT-6 Astra
OpenAI
- Plan: Pro
Gemini 3.8 Flash
Google
- Plan: Pro
Grok 4.6
xAI
- Plan: Pro
DeepSeek V4 Pro
DeepSeek
- Plan: Pro
Qwen 3.8 Max
Alibaba Qwen
- Plan: Pro
Mistral Medium 3.5
Mistral AI
- Plan: Pro
GLM 5.3
Z.ai
- Plan: Pro
LLaMA 4 Scout
Meta AI
- Plan: Free
Nemotron 3 Ultra
NVIDIA
The tag shows the lowest plan that includes the model.
See measured comparisonsBlind Prompting
Names and numbers are masked before the model sees them.
With Blind Prompting on (Settings), personal data is swapped for tokens before it leaves your browser. When the answer comes back, the tokens are restored only on your device.
You wrote
Asilbek Yusupov, FayzInc, $50,000 budget
Masked in your browser
AI sees
[PERSON_A], [ORG_A], [VAL_1] budget
Pricing
Start for free. Upgrade when you need to.
Free plan: Llama 3.3 70B, Gemini 2.0 Flash and DeepSeek R1, up to 30 messages a day. Paid plans add flagship models, full code generation and web research with sources.
Visa/Mastercard · USDT/BTC · SBP (₽)
Free
$0/mo
To try it out
Sign up and get instant access to free models.
- Free models: Llama 3.3 70B, Gemini 2.0 Flash, DeepSeek R1, Nemotron 3
- With a daily limit
- Medium-length code and answers (3K tokens)
- Chat history
- Memory
Basic
$5.99/mo
≈ $0.20/day
For everyday use
3× more tokens, with affordable flagship models.
- Everything in Free
- Adds Claude Haiku 4.5, GPT-5.6 Luna and Gemini 3.1 Flash Lite
- 3× more tokens / month
- Longer answers (4K tokens)
- Code: complete medium-sized examples
- Answers with thinking — the model's reasoning is shown
Popular
Pro
$21.99/mo
≈ $0.73/day
For pro-level work
10× more tokens, all flagship models and research.
- Everything in Basic
- Claude Sonnet 5, GPT-5.6 Sol, Gemini 3.5 Flash, Grok 4.6, DeepSeek V4 Pro
- Perplexity Research (with sources)
- 10× more tokens / month
- Full code generation (4K tokens)
- Full CLI coding agent
- Answers with thinking — routing prefers reasoning models
Ultra
$109.99/mo
≈ $3.67/day
Maximum power
20× more tokens, Claude Opus 5 / GPT-6 Astra, deep research, priority queue.
- Everything in Pro
- Claude Opus 5 and GPT-6 Astra flagship models
- Perplexity Sonar Pro (deep research)
- 20× more tokens / month
- Longest answers (8K tokens)
- Priority queue
- Early access to new models
- Answers with thinking — the strongest reasoning models
Roadmap
What's live, and what's next.
What ships today and what we're building next.
Shipped
Live- Web app
- CLI code agent
- Desktop app
- Image generation
- Blind Prompting (personal data masking)
- 4 interface languages
- Card, crypto and SBP payments
- Agents and Telegram bot
- Crownwork Mods and gallery
- Sovereign Mode (offline local agents)
- UzBench — Uzbek-language benchmark
Next
Planned- Video generation
- Music generation
- Tella — our own fine-tuned model
- Mobile app
FAQ
Frequently asked questions
Short, honest answers. Details live in the docs and the privacy policy.
What is SOVEREIGN?
An Uzbek-first AI workspace: Claude, GPT, Gemini, DeepSeek, Qwen and more in one account; Auto mode picks the model for each question. Crownwork is a desktop agent that works with files on your computer, the CLI runs in the terminal, and scheduled agents deliver results to Telegram.
Which AI understands Uzbek best?
Measurement, not guesswork: UzBench is an open benchmark for the Uzbek language (math, language, knowledge, reasoning, dialects). The leaderboard is at /uzbench; questions are public and grading is automatic. Inside SOVEREIGN, Auto mode picks the right model per request.
Does it understand Uzbek dialects?
Yes. When a message shows dialect features (Fergana “kevotti”, Kashkadarya “jaxshi”, Khorezm “galdi”, Tashkent “vomman”), the model receives a short guide covering 12 regions: it doesn't “correct” you, it understands the meaning and answers in standard Uzbek (or in your dialect if you ask). This works for every model; UzBench has a dedicated “Dialects” section.
Is my data safe?
Honestly: a request goes to the provider of the model you chose — there's no way around that. What we do: Blind Prompting masks personal data before sending; in Crownwork, Sovereign Redact masks keys, passwords, emails, phone and card numbers on your computer and restores them in the answer; Sovereign Mode agents run fully offline with Ollama. Chats are never used for ads or sold.
How much does it cost and how do I pay?
There is a Free plan; Basic, Pro and Ultra are monthly subscriptions, with two months free on annual billing. Pay by bank card, crypto or SBP. Prices are on /pricing.
What is Crownwork and how is it different from Cursor or Claude Code?
Crownwork is a desktop AI agent for Windows, macOS and Linux: it reads and writes files, runs commands and asks for approval on every action. The difference: multi-model (no single-vendor lock-in), open — you reshape its interface through Mods and Studio, Sovereign Redact masks secrets on your computer, and Sovereign Mode works offline. Download it from the “Download” section on this site.
About
Independent. Privacy-first.
Our mission is to make the world's best AI useful and affordable for people in Uzbekistan and the CIS — in their own language, with payment methods that work locally, and without handing personal data to AI companies. SOVEREIGN is designed, built and operated independently.
- Company
- SOVEREIGN AI
- Location
- Tashkent, Uzbekistan
- hello@soveregn.xyz
- Website
- soveregn.xyz
Team
Islombek
Founder & developer
Designs and builds SOVEREIGN end to end — web app, CLI, desktop app, payments and infrastructure.
Contact
Questions? Write to us.
Support, billing, partnerships or press — email us and the founder reads every message.