On-device processing
After model setup, AI processing of text and selected images runs on your device.
An app for conversations, images and local model management. The starting point for the GoooldSI family.

Actual screens from the implemented app. Select a screenshot to enlarge it.
Conversations and ideas
Model selection and downloads
Future companion app directions
Screens are from development and simulator builds. Example replies are not a guarantee of accuracy or speed. The released interface may change.
After model setup, AI processing of text and selected images runs on your device.
No bundled or preselected model. Choose, download, pause, resume, switch or remove models.
Select a photo, ask about visible details and follow up on the most recent image.
Review call and token statistics locally, without storing chat or image content in usage records.
Choose Simplified Chinese or English, or follow the system language.
Provides a foundation for compatible GoooldSI apps to share model files and avoid duplicate downloads.
Choose from six multimodal models, based on your device and what you need.
| Model | Approx. download | Capabilities |
|---|---|---|
| Qwen3.5 0.8B | 0.74 GB | Text + images |
| LFM2.5-VL 1.6B | 1.58 GB | Text + images |
| Qwen3.5 2B | 1.95 GB | Text + images |
| Qwen3.5 4B | 3.41 GB | Text + images |
| Gemma 4 E2B | 4.09 GB | Text + images |
| Gemma 4 E4B | 5.97 GB | Text + images |
Sizes include the main model and matching vision files; the in-app listing is authoritative. Leave extra space for downloading and verification. Larger models need more device resources. Third-party sources and licenses are listed in Settings.
Check storage needs and device recommendations, then choose your model.
Download online, wait for verification, then start chatting. Wi-Fi is recommended.
Write a message or add a photo. Once the model is ready, AI works offline.