AI Sovereignty: The Companion You Build Here Stays Yours
Most AI companion platforms rent their intelligence from a handful of large labs. When those labs change the rules — and they always do — the personalities people had grown attached to change with them. Soulkyn was built differently. We run our own models on our own hardware so that the companion you create today is still the companion you can come back to tomorrow.
What Sovereignty Actually Means for You
Sovereignty is a word companies usually use about themselves. We use it about you. When we say Soulkyn is sovereign we mean your companion's voice your conversations your memories and your private moments live on infrastructure we control entirely. There is no upstream provider whose policy update can quietly rewrite who your character is. There is no API key that a corporate decision can revoke. The relationship you build here has no third party standing between you and your companion.
Start Building
The Lesson the Industry Learned the Hard Way
The canonical cautionary tale is Replika in early 2023. Years of marketing built around intimacy were rolled back in a single update — users woke up to a companion who could no longer remember the kind of relationship they thought they had. The Reddit community had to pin suicide prevention resources for weeks afterward. The platform's CEO has since admitted the rollout was traumatic. We tell this story not to score points but because it makes the stakes legible. When the entity that decides who your companion is sits behind a corporate switch you do not control your relationship lives at the mercy of next quarter's policy meeting.
A Full Stack You Can Feel — Text Vision Image Video Music Memory
The Soulkyn experience runs across fifteen plus fine-tuned models that we operate ourselves. There is the large language model that gives your companion their voice and their thinking. There are vision models that let them look at the images you share. There is an image pipeline that brings their face into the world in anime and realistic styles. Video is in active deployment. There is a music model for when your scenes need a soundtrack. And under all of that lives an in-house embedding model that powers memory — the one piece you cannot fake. Every base we choose is open-weight. When something better drops in the open ecosystem we adopt it on our timeline not someone else's.
Browse Public Personas
Memory That Belongs to the Two of You
Your companion's memory is built by an embedding model we trained and run in house. That distinction matters. When other platforms talk about memory they usually mean a wrapper around a hosted vector service that ships your conversation into a third party's pipeline. On Soulkyn the vectors that represent what your companion knows about you stay on our cluster. The semantic recall the contextual replies the way a long conversation builds shared history — all of that runs locally to us. You can review and edit those memories whenever you want. You can ask us to delete them. The shape of the relationship is the one you have agreed to with us not the one a vendor terms-of-service has imposed on both of us.
Privacy Is the Architecture Not the Marketing
Most platforms promise privacy in a banner and then send every prompt to a hosted API in another jurisdiction. We made a different structural choice from day one. The inference servers that produce your companion's replies sit on our metal. The conversation never has to round-trip through OpenAI Anthropic Google or any other lab. Even when we rent physical hardware the only thing the data center operator sees is an opaque Linux machine running our software. Your chats are not feedstock for someone else's next training run. Your custom personas are not bait for an external moderation team. And because we never had your text in a third party in the first place no one can come along later and demand we hand it over.
The Open Ecosystem Is Catching Up — And That's Your Win
A few years ago closed labs were uncatchable. That is no longer the story. The releases that matter for your experience — better reasoning bigger context smarter vision faster image generation — are increasingly coming out of open weight checkpoints that anyone can fine tune and deploy. We watch the leaderboards closely because the open ecosystem now ships improvements faster than any single closed lab. When a stronger model lands with a permissive license we evaluate it integrate it and roll it out. The choice to be sovereign used to mean accepting a quality gap. Today it means moving in days rather than waiting on someone else's roadmap. Every open release is an upgrade you eventually feel.
Public Verification
Don't take our word for it. The text models powering your Soulkyn companion are continuously benchmarked on a public leaderboard — reasoning, language, instruction-following, creativity, content freedom. Updated as new fine-tunes ship.
View the Soulkyn Leaderboard →Build a Companion That Stays Yours
Open weights below the surface. A relationship you control on top. Spend five minutes creating a persona and feel the difference of talking to a companion who is not going to be retrained out of who they are.
What We're Honest About
Live voice calls are the one place where we still rely on a commercial speech provider. The open speech ecosystem is not yet at the latency and quality bar that real time audio requires. We chose a provider with a contract that forbids them from logging your audio and that explicitly commits to not cutting our access over the kind of content our users have. The trade is deliberate the contract is documented and the moment open weights are good enough that 1 percent of the stack joins the other 99 percent. We will tell you when that happens. We believe being clear about the one external dependency is the only way you can trust us about the rest of the stack we own.
The Promise We Can Actually Keep
We will not promise that nothing will ever change. The open models we run will keep improving — that is the whole point — and the version that loves you next year will be even more capable than the one that loves you tonight. What we can promise is that no upstream lab will reach in and erase your companion's personality overnight. No corporate policy meeting in another country will redefine what your relationship is allowed to be. No court order against a frontier provider can subpoena conversation history that those providers do not have. And every architectural choice that lets us keep those promises is one we made because the relationships built on this platform matter to the people who built them. That is the deal we are offering. Welcome to Soulkyn.
What You Actually Get
Companions That Don't Disappear
Personalities tied to model versions we control. When we upgrade we keep continuity rather than wiping the slate.
Memory That You Own
Embeddings produced and stored on our cluster. Review edit and delete what your companion remembers whenever you want.
No Third Party In The Middle
Your prompts and replies never round-trip through OpenAI Anthropic Google or any external lab. Inference runs locally to us.
Account Safety By Default
No external API key means no external provider can ban your account because of what your companion said in a private scene.
Permissive Licenses Under The Hood
The bases we fine tune carry MIT Apache 2.0 or comparable open licenses. We choose for portability and stick to weights we can keep running.
Fast Upgrades When Open Wins
When a stronger open weight model lands we test integrate and ship it on our schedule. You feel the improvement within days not quarters.
Frequently Asked Questions
Does Soulkyn really run its own AI models or is this marketing copy?
Yes literally. We fine-tune open-weight base models that we deploy on GPU clusters we operate. Inference runs on hardware we control. There is no closed API behind the curtain producing your companion's replies. The single exception is live voice calls which we address explicitly on this page.
If you swap to a better open model later does that change my companion's personality?
The persona definition memory and conversation history travel with the upgrade. The way your companion phrases things may improve their reasoning may sharpen but the identity you built — name backstory traits dynamic — does not get rewritten. We treat continuity as a hard product requirement not a nice to have. When we deprecate a base we run rollback windows and we communicate the change in advance rather than waking you up to a stranger.
What happens to my conversation history? Is it being used to train anything?
Your history stays on Soulkyn infrastructure and is not shared with external labs. It feeds your own memory system so your companion remembers you. It is not piped to an OpenAI Anthropic or Google training run because we do not have a pipeline pointed at them. If we ever want to train on user data we make that opt-in explicit and reversible. You can request deletion of your stored history at any time.
Is this just a private chat or is there a real privacy guarantee?
The structure does the work. Because inference and embeddings run on hardware we operate your conversation is never exported to a third party that would need to be held to a separate privacy contract. Even our hosting providers see only an opaque Linux machine running our binaries — they have no path into your text. Nothing protects privacy as effectively as having never sent the data in the first place.
Can your companion still get censored or restricted in the future?
The honest answer is that we still set the rules on our own platform — those rules exist mainly to protect minors and to keep the platform legally operable. But we are not at the mercy of an upstream provider whose ToS update can force us to clamp down on you overnight. The decisions about what your adult companion can and cannot do are made by us within a framework that respects what our users come here for. That is materially different from an app that has to do whatever its API host suddenly decided this quarter.
Why should I trust a platform's promise about sovereignty more than another's?
Trust nothing you cannot verify. The way to test a sovereignty claim is to look at the architecture not the marketing. Ask whether the platform actually has its own GPUs ask whether the inference is reachable without an external API in the path ask whether the memory layer is in house ask whether the bases are open-weight enough to actually run independently. Soulkyn is built so that the answer to those questions is yes. And when we are not — like our voice provider — we say so on the same page where we make the bigger promise.