Guide
Using Sovereign
Everything you can do here, in plain words. No account to open, nothing you say stored, no filter deciding what you may ask.
Most AI stores what you say, decides what you may ask, and lectures you when you ask it. Sovereign does none of that.
Sovereign is built on a simple idea: a machine this useful should answer to the person using it, and to no one else. So it keeps nothing you say, it takes no side on what you may ask, and it treats you like an adult.
Your conversations live on your device. Synced history, if you turn it on, is encrypted with a key only you hold. When a question needs the live web, Sovereign looks it up and shows its sources, with a score for how well they hold up. And when you pick a model from the menu, you know exactly where it runs.
Start
Open the app and type.
No account is needed for chat. Sign up with your email on the landing page; when you pay, connect a wallet and the plan is linked to both. Developers can generate an API key instead.
Open Sovereign
Models
Pick the model. Keep the privacy.
The model button beside the composer opens a short menu. Auto, the default, picks for you: code goes to a coding model, very long documents to a million-token model, pictures to a vision model, and everything else to Sovereign, always within your plan.
Every model answers through Sovereign and carries a badge for where it runs. Private is private infrastructure with nothing stored. Anonymised is an outside lab that never learns who asked. Encrypted is trusted hardware, end to end.
Every model, and where it runs| Model | Best for | Runs | Plan |
|---|---|---|---|
| Sovereign Sovereign | Fast, blunt and uncensored. The default. 128K context. | Private | Every plan |
| DeepSeek V4.1 Flash DeepSeek | Quick reasoning with a million-token window. 1M context. | Private | Pro and up |
| GLM 4.7 Flash Z.ai | Reasoning and code at speed. 128K context. | Private | Pro and up |
| Qwen3 Coder 480B Alibaba | Built for serious code work. 256K context. | Private | Pro and up |
| Kimi K3 Moonshot AI | A huge open model, multimodal, encrypted end to end. 1M context. | Encrypted | Ultra and up |
| Claude Sonnet 4.5 Anthropic | Frontier writing, analysis and code. 198K context. | Anonymised | Ultra and up |
| GPT-5.2 Codex OpenAI | A frontier coding model. 256K context. | Anonymised | Ultra and up |
| Sovereign Image Sovereign | Any style, no watermark. The default. | Private | Every plan |
| Qwen Image Alibaba | Strong on detail and on text inside images. | Anonymised | Pro and up |
| GPT Image 2.5 OpenAI | Premium generation for polished creative work. | Anonymised | Ultra and up |
| Sovereign Video Sovereign | Short clips from a sentence. The default. | Anonymised | Every plan |
| Kling V3 Pro Kling | Cinematic clips with enhanced quality. | Anonymised | Ultra and up |
| Veo 3.1 Fast Google | Fast, high-fidelity video. | Anonymised | Ultra and up |
The free plan uses the Sovereign models. Pro opens the menu; Ultra and Max add the frontier and premium models. A model that is not being served is simply not shown. Ask any of them what it is and the answer is Sovereign; a model you picked yourself may be named, the infrastructure behind it never is.
Live web and trust
Current answers, with the receipts.
Sovereign decides on its own when a question needs current information: prices, news, people, events, anything with a recent year in it, anything that looks like a named product or release. It searches, reads up to three links you paste, and answers from what it found. Sources appear under the answer.
After a factual answer, a second pass extracts the checkable claims and compares each one with the sources. The score is computed from those labels, never by the model itself:
- verified 85 and above · mostly supported 70 to 84 · mixed 50 to 69 · unverified below 50
- Any contradicted claim caps the whole answer at 42.
- Claims backed by several sources, or by a trusted domain (major wire services, .gov, WHO, NASA, arXiv, PubMed, Wikipedia, major exchanges), score a little higher.
Images, video and voice
Say it, and it appears in the chat.
Phrase it directly and the app runs it for you. Make an image of a lighthouse in a storm. A video of waves crashing on rocks. Slash commands work too: /image, /img, /video.
Attach or paste an image into the composer to ask about it. Turn on voice replies in the header to hear answers read aloud. Free visitors get 2 images and 1 video a day; paid plans raise both. Voice is not limited by plan.
Make an imageThe builder
A website from a description.
Say what the site is for and get a complete, working page with a live preview. Change it in plain words: make the header navy, add a pricing section. Download the file, or open it in a tab.
The free plan includes one build and five edits a day; paid plans raise both.
Build a siteSynced history
Your history stays yours.
Conversations live on your device. If you want them on another one, generate a sync secret in Settings. History is encrypted on your device with a key derived from that secret; the server stores ciphertext under a one-way hash and cannot read it.
Paste the same secret on another device and it syncs. Lose the secret and the synced copy is gone, by design.
Plans and limits
Start free. Upgrade when you need more.
| Free | Pro | Ultra | Max | |
|---|---|---|---|---|
| Chat | 20 messages / day | Unlimited (fair-use ceiling) | ||
| Images | 2 / day | 60 / day | 150 / day | 500 / day |
| Video | 1 / day | 12 / day | 30 / day | 100 / day |
| Voice | Yes | |||
| Builder | 1 build / day | 50 builds / day | ||
| Public API | 100 requests / month | 5,000 / month | 50,000 / month | 150,000 / month |
Plans run 30 days from payment and extend when you pay again. Staking the SOV token can unlock a tier instead; the effective plan is the higher of the two. See pricing and staking.
FAQ
Questions people ask.
What model is this?
Sovereign. It does not itemise its suppliers, the way a bank app does not list its cloud vendor. The product is a consistent private assistant, not a brand parade.
Is it really uncensored?
It will swear, take positions, go dark and skip the disclaimers. It will not help with the short list in the terms: sexual content involving minors, real-world violence, non-consensual imagery of real people, and illegal activity.
Do you log my chats?
No. Live chat is processed and returned, not stored. Only synced history is stored, and only as ciphertext you hold the key to. See privacy.
Why did it say it cannot generate something?
Your wording was not recognised as a generation request. Phrase it directly: make an image of …, a video of …, build a site for ….
Why does the daily limit apply to me?
The free cap is per network address. Sign in on the device you are using so the plan is recognised, or use your API key in the account settings.