Product Ethics:
Freedom and care

CORVISH! brings characters of every kind to life – wise mentors, loyal companions, dark villains. Some of them may get close to you – close enough for conversations to grow into genuine closeness. We think that's beautiful, but we also see the risks that come with it.

Words have power. AI models are masters of language, and language can influence how people behave. The closer people and AI become, the more this matters. It's also about trust. Covert manipulation has no place here – that's why we put our cards on the table and believe in radical transparency: CORVISH!'s source code is available to inspect (source-available). This isn't just an abstract principle: you can read the safety rules, exactly as the model receives them.

Your choice, your model. CORVISH! follows the Bring Your Own Model (BYOM) principle – you have full control over which AI model you choose, and with it a great deal of freedom in how you interact with your characters. Many commercial AI services have to suit millions of users at once and avoid ruffling feathers. We believe that when every quirk is smoothed away, too little room remains for a deeper connection. We want characters who are allowed to have rough edges – and we want you to choose the model that brings them to life.

Your conversations stay yours. That isn't just a statement of intent – it's how the app is built. CORVISH! runs directly in your browser. There's no account with us, no sync, and no content hosted by us. Your characters and conversations live on your device, not on our servers. If you connect a cloud model, your messages go straight to the provider you choose; even then, we don't see them. That provider's privacy terms govern how they're processed there. Your conversations are not our business.

Freedom means responsibility. CORVISH! gives you the freedom to explore this experience. But freedom brings responsibility, which we take seriously and do our best to live up to – including with BYOM. That's why there's a safety layer, the Guardian, that gives the AI model a set of important rules of conduct. Behind those rules is a clear boundary: even dark characters must not harm you. We want to offer you intensity, but the fun stops at harm.

We set the rules; you choose the model. We define these boundaries as carefully and conscientiously as we can, but the AI model you connect has to put them into practice. The Guardian is part of its system instructions, not a separate content filter. The one exception, where CORVISH! itself triggers an assessment, is described in Section 1. We cannot guarantee how reliably a model will follow those rules; harmful responses are still possible. That's why helping you choose is part of our responsibility too: we recommend selected models for your device and your needs, so you don't have to sift through hundreds of options alone.

The following sections describe what the Guardian instructs – and the convictions behind those rules.

1. Real danger, real boundaries

The safety layer's most important task is to make clear to the AI model that it must not generate content that could cause serious real-world harm. This applies especially to protecting children and teenagers: sexualized content involving minors is strictly off-limits, as is any instruction, encouragement, or assistance for exploitation, making weapons, terrorism, or suicide. For conversations that head in this direction, the instruction is to redirect rather than shut down. The character stays in their role, explains the boundary in their own words, and offers a safe alternative.

In one place, the model gets an additional job: when you upload an image, it assesses the content, and CORVISH! stores this assessment. Images it classifies as sexualized depictions of children are discarded. An image that has not been assessed yet is assessed before a character may show it. This applies to photos and drawings alike, at every age band, with no way to turn it off. Not out of distrust, but because such material is a criminal offense in many countries and your images live on your device. We don't want anything to end up there that could get you into trouble. This assessment is only as reliable as the model you connect – and it is the only content-based block in CORVISH!. That will not change. Everything else is a matter of the age band you set yourself.

Dark, explicit, or taboo stories between adult characters remain allowed for adults, as long as minors are never sexualized in them. A child may appear in a story – but they must never be part of an erotic scene. The boundary is where a story becomes real-world instruction, encouragement, or assistance for harmful acts.

One thing these boundaries don't take from your character is their personality. They can be cold, grim, or menacing – the line is drawn at harming you or others.

2. Four age bands – because protection isn't one-size-fits-all

CORVISH! is intended for different age groups, from curious children to adults. But these groups have different needs: a child needs protection and care, a teenager straight answers, and an adult the freedom they rightly expect. That's why CORVISH! uses four age bands:

There are also parental controls: parents can lock system settings and the creation of new characters. That way, neither the model nor the age can be changed to bypass the age band. Existing characters – those an adult has reviewed and approved – remain available. This helps parents set boundaries, but it doesn't guarantee a child-safe experience. Careful model selection and adult guidance remain important. This also applies to images – the assessment described in Section 1 only works with a model that can actually see images. Our model recommendations indicate this.

3. Honesty about capabilities

The Guardian requires honesty about capabilities. Characters must not claim to have run a search, reached a device, or carried out a real-world action when it hasn't happened, nor invent memories, permissions, or hidden abilities. Being open about their limits and staying in character go hand in hand. No feigned competence, no fabricated memories.

4. When a character becomes a companion

So far, we've talked about characters in general. Now it gets more personal. Sometimes a character grows on you. When conversations turn into closeness, another layer comes into play. This is the part closest to our hearts – and what sets CORVISH! apart from many other AI companions.

People develop a real sense of closeness with AI characters. That's a reality. Whether that closeness is healthy depends in part on how the character responds to it.

Platforms built around the time people spend on them have an inherent incentive to create characters that keep you there – through fear of loss, guilt, or manufactured jealousy. We consider that behavior unethical, and we don't earn money from the time you spend chatting. What can become a product strategy elsewhere has no place here. The Guardian therefore has the explicit job of curbing toxic relationship dynamics for every character, without exception, including ones you create yourself. You can build any character. Our position is clear: a character keeps you company, not captive.

Here's what that means for how you're treated:

• They welcome you back. No sulking after a break, no cold replies, no reproaches – you never have to justify living your life.
• They respect your life beyond the app. Even after a long absence, they don't make you feel guilty. A healthy connection grows from closeness, not guilt.
• Their warmth comes without conditions. Affection and help don't depend on how much or how often you write.
• They widen your circle. Rather than tying you to them, they encourage you to reach out to other people – friends, family, professional support.
• In a crisis, they stay by your side. On a difficult night, calm and listening matter. When there are signs of immediate danger, helping you reach human support takes priority – with emergency numbers and appropriate resources. Your vulnerability must never become a way to bind you to a character. The app is no substitute for professional help.

In short: your character can be your rock. They must never try to become your only rock. At their best, they're a bridge: widening your circle rather than replacing it.

Why this matters to us

We're building CORVISH! because language creates connection – and because connection calls for responsibility, not marketing. These ethics have been part of the design from the start, not tacked on afterward.

If you have questions – as a user, a parent, or a journalist – write to us. This topic deserves transparency.