내가 당신이 보여주는 것을 볼 수 있는 iOS AI 컴패니언을 구축한 방법

Most AI companions only know what you type or say. Scowld adds an optional third input: a single camera frame you deliberately choose to share.

The interaction I wanted

Describing an object is often harder than showing it. With vision enabled, you can hold something up, point the camera at a screen or show the room, then ask a question naturally. Scowld sends that one frame with your message to the vision-capable AI provider you selected and reads the reply aloud.

This is not continuous recording. Vision is off by default, a frame is captured only when you send a message with the camera enabled, and Scowld does not save that image to the photo library.

Why the provider choice matters

Scowld is bring-your-own-key. You can use a vision-capable provider such as Gemini, OpenAI or Claude, and the request goes from the device to that provider. There is no Scowld server proxying the frame or storing your API key.

More than a camera demo

The rest of the conversation still works by voice or text. The companion has animated reactions, multiple characters, configurable personalities and separate locally stored chats. Vision is just another optional way to give the conversation useful context.

Try or inspect it

• App Store: https://apps.apple.com/app/id6760672848
• Source code: https://github.com/apoorvdarshan/scowld
• Technical overview: https://scowld.xyz/blog/ai-companion-that-can-see

Scowld is available on iPhone and iPad and released under the MIT License.