SOVEREIGN AI docs

Create an account at soveregn.xyz/register with your email, or continue with Google or GitHub. New accounts start on the Free plan — no card required. A short onboarding runs once after sign-up, then you land in the app.

• Open the app at app.soveregn.xyz and type a message. Enter sends; use the paperclip to attach images, PDFs, text or code.
• Pick a model from the model switcher, or leave it on Auto (the default).
• Your conversations are saved in the sidebar so you can come back to them.
• The + menu next to the input holds extra tools: attach a file, your Knowledge base, Memory, Create image and the Cowork folder.

The default model is SOVEREIGN Auto. Instead of you choosing a model, the server looks at the request first and routes it to the model best suited for the task — or to a chain such as research → code — within what your plan allows. You can always override it by choosing a specific model in the switcher. Models your plan does not include are shown with the plan that unlocks them.

Open the + menu and choose Create image, then describe the picture. Generation takes about 30 seconds. Daily limits by plan: Free 3, Basic 15, Pro 60, Ultra 200. The counter resets every 24 hours. If Blind Prompting is on, the image prompt is masked too.

Toggle Research mode under the input to answer with live web research via Perplexity, with numbered citations and a sources panel. Research is available on Pro and Ultra; deep research (Perplexity Sonar Pro) is Ultra-only. The Researcher and Business analysis agent modes turn research on automatically.

SOVEREIGN remembers key facts — your name, projects and preferred style — across chats. Open Memory from the + menu to see what it knows, delete single items or clear everything. See Privacy for how memory is stored.

Upload your own documents — PDF, text, Markdown, JSON, CSV or code — in Knowledge base. They are indexed so the AI can answer from them. Reference a document in a message by typing @ and its name; relevant passages are also retrieved automatically.

The Connectors panel lets you link outside services so the AI can work with them:

• Google — Gmail, Drive, Sheets, Slides, Docs and Calendar (Google sign-in).
• Design & dev — Figma, GitHub, or any MCP server (URL plus optional token).
• Built-in — CLI terminal (runs commands through sov on your computer, with a risk check), Browser preview, and keyless public APIs (weather, currency, countries, crypto, time, dictionary).

Each connector can be switched on or off at any time.

SOVEREIGN Skills are expert playbooks added to the model's instructions — for example UI/UX Pro Max, Apple Design, Clean Code, Cybersecurity Pro, Pro Writing and Data Viz. A skill switches on automatically when your message matches it, or you can toggle it yourself.

Agent mode sets the overall behaviour: General, Developer, Researcher, Business analysis or Writer.

Switch the input from Chat to Cowork and choose a folder on your computer. SOVEREIGN can then read files from it directly — pick files with @ instead of uploading them. The folder itself is never uploaded: only the files that are actually opened are sent, and your browser enforces the permission you granted. For a full desktop agent that writes files, see Desktop app and CLI.

Turn on Private mode under the input. Before your message is sent, personal data is replaced in your browser with placeholders such as [PERSON_A] or [EMAIL_A]; the model only ever sees the placeholders, and the real values are put back into the answer on your device. The same value always gets the same placeholder within a conversation. Masked categories:

• Names of people (Latin, Cyrillic and other scripts) and company names (LLC, Inc, MChJ, GmbH…)
• Email addresses, phone numbers and URLs
• Card numbers, IBANs, amounts of money and crypto wallet addresses (BTC, ETH, SOL, TRX)
• IP addresses and ID numbers (passport, SSN, TIN formats)

Masking is pattern-based, so it can miss unusual formats — avoid pasting secrets you would not share.

Click the microphone in the input to dictate. Speech is recognised by your browser in your interface language; the button only appears in browsers that support speech recognition (for example Chrome and Edge). Transcribing attached audio files is available on Pro and above.

Requires Node.js 20+ with npm. Pick your shell:

$ irm https://soveregn.xyz/install.ps1 | iex

$ powershell -c "irm https://soveregn.xyz/install.ps1 | iex"

$ curl -fsSL https://soveregn.xyz/install.sh | sh

$ npm install -g @islombekrrr/sov-cli

Then start it:

$ sov

The installers check your Node.js version and run the npm install for you (on Linux they retry with sudo if the global folder needs admin rights). Both sov and sovereign are installed. Uninstall with npm uninstall -g @islombekrrr/sov-cli.

On first run, choose how to connect: your SOVEREIGN account (recommended — a browser window opens, you click Allow, and your plan applies) or your own OpenRouter key.

sov                               # interactive mode (chat + agent)
sov "build a React todo app"      # one task, then exit
sov "..." -f diagram.png -f a.pdf # attach files (--file)
sov --yes "..."                   # auto-confirm safe actions (-y)
sov --full-auto "..."             # Full auto: nothing asked (push/deploy refused)
sov login                         # connect your account in the browser
sov logout                        # sign out
sov whoami                        # connection status
sov key sk-or-v1-...              # use your own OpenRouter key
sov init --ai                     # SOVEREIGN.md: shared project memory for the team
sov audit [--json]                # pre-deploy security audit (offline)
sov models                        # list models
sov help                          # help

Attachments: images (.png .jpg .webp .gif, up to 900 KB), PDF (up to 20 MB) and text/code files (up to 2 MB). Credentials are stored only on your machine in ~/.sovereign/config.json; you can also set SOVEREIGN_TOKEN or OPENROUTER_API_KEY.

On Free and Basic the CLI uses Auto routing; choosing a specific model with /model needs Pro or higher. The CLI has its own daily step limit based on your plan.

Type / in interactive mode to open the menu.

So no AI session starts from zero: the project overview, stack, commands (test/build/run), rules and a “don't touch” list live in one SOVEREIGN.md file. The agent (CLI and Cowork) reads it at the start of every conversation. Commit it to git — your teammates' agents start from the same rules and stop re-creating functions that already exist.

sov init          # SOVEREIGN.md template
sov init --ai     # let the agent fill it in
git add SOVEREIGN.md && git commit -m "Project memory"

sov init creates a template (stack and commands are guessed from package.json); sov init --ai lets the agent study the project and fill it in (it asks before writing). The file is looked up from the current folder to the git root; .sovereign/PROJECT.md works too (if both exist they are merged). Limit: 16 KB.

Add a rule: /project-remember <fact> in the CLI (view with /project), or right panel → Project → “Remember” in Cowork. It is appended with a date under ## Eslatmalar. When you say “always do X” or “don't touch Y”, the agent suggests saving it — it never writes on its own.

sov audit checks your project offline — no files leave your machine and no AI is involved. It catches the most common holes in AI-built apps before you deploy. Keys are never printed in full: only the first 4 characters + ***.

$ sov audit

$ sov audit --json > audit.json

It checks: keys in code (AWS, Stripe sk_live, OpenAI, GitHub, Supabase service_role, private key files); .env files that are not gitignored or already committed; secrets in NEXT_PUBLIC_ / VITE_ / EXPO_PUBLIC_ vars; server env and service_role in "use client" components; Supabase tables without RLS and using (true) policies; open Firebase rules; CORS * + credentials; eval, dangerouslySetInnerHTML and http:// API URLs.

• File actions are limited to the folder where you started the CLI.
• Before writing a file, creating a folder or running a command, it asks for confirmation (y / N, or a to allow in-folder actions for the rest of the session).
• Vibe mode is on when you start it as sov (or with --vibe; toggle with /vibe): file changes inside the working folder and read-only commands run without asking. Start it as sovereign to confirm every change.
• Full auto (--full-auto or /auto; the ⚡ button in Cowork) — nothing is asked: file writes, package installs, tests and builds inside the working folder run right away, and the agent keeps fixing until tests pass. Paths outside the working folder, git push, publish, deploy and sudo are refused without asking; protected paths and dangerous commands stay blocked. Turn it on only in a folder you trust.

Have a promo code? Enter it in the plan dialog — it works with every payment method. Each account can use a given code once, and codes can have a limited number of uses.

Every payment is treated as value. You always stay on the highest active plan, and time is converted between plans by their monthly price ratio:

• Upgrading — the time left on your current plan is converted to the new plan (e.g. 10 days of Basic at $5.99 becomes about 2.7 days of Pro at $21.99), then the new period is added on top.
• Buying a lower plan while a higher one is active is blocked — extend your current plan, or buy the lower one after it ends.
• Card subscription active? To change plans, cancel it first (link in your Dodo email) or pay by crypto/SBP — your remaining time is converted automatically, so you are never charged twice.

• Cancel anytime — no further payment is taken and access continues until the end of the paid period.
• 7-day refund on the first payment of a paid plan if you have not used the service significantly.
• Not refundable: renewals after the 7-day window, periods with significant use, and accounts suspended for breaking the terms. Used quota is not grounds for a partial refund.
• Crypto payments may be non-refundable due to the nature of blockchain; reviewed case by case.
• Email support with your account email and payment date; requests are usually handled in 3–5 business days.

Full policy: soveregn.xyz/refund.