1. Local AI is separate from remote processing
When a model is installed, OmniPrism Studio uses it on your computer for the feature that requires it. Your document content, extracted text, recordings, images, embeddings, and search context are not sent to a remote AI API by default.
Model downloads are different: the hosting service may receive an IP address, user agent, request time, and download information. The download request does not contain your library content.
2. Current model families
The exact quantization, revision, checksum, and enabled capability are release-specific. The application manifest and the bundled release notice are authoritative for the files used by the current build.
| Family or component | Use in OmniPrism Studio | Upstream licence | Upstream source |
|---|---|---|---|
| BGE-M3 | Dense embeddings and semantic retrieval | MIT | BAAI model page |
| BGE reranker v2 M3 | Optional result refinement | Apache-2.0 | BAAI model page |
| Qwen3 and Qwen3-VL | Optional local query interpretation and vision-assisted processing | Apache-2.0 | Qwen model organisation |
| Whisper / whisper.cpp models | Optional local audio transcription | MIT for whisper.cpp; verify model-specific notices | whisper.cpp model repository |
For release 1.0.0+1, the bundled notice records the exact model files, revisions where pinned, byte sizes, and SHA-256 values. Qwen model URLs currently use upstream main; the recorded checksums must be reverified whenever the model configuration changes.
3. Licence and attribution notices
Each redistributed model, tokenizer, adapter, and runtime is subject to its applicable upstream licence and notice requirements. The release package should include the relevant licence texts and attribution notices for the exact components it distributes.
Upstream model cards and repositories are linked for reference. If an upstream licence, attribution requirement, or redistribution condition changes, the release inventory must be reviewed before shipping an updated model.
Model attribution is an intellectual-property and redistribution requirement. It does not by itself mean that the model provider receives your documents or search queries.
4. Runtime and open-source components
OmniPrism Studio may use ONNX Runtime, llama.cpp-compatible components, whisper.cpp components, SentencePiece/tokenizer assets, and other open-source libraries. The exact component and licence inventory for release 1.0.0+1 is available in the bundled release notice.
For a copy of the notices or a question about attribution, contact [email protected].
5. Accuracy and responsible use
OCR, transcription, query interpretation, embeddings, and reranking can produce errors. OmniPrism Studio is designed to help you locate and inspect original evidence; it does not guarantee that an AI-derived result is complete, correct, or suitable for a legal, medical, financial, employment, or other high-impact decision.