OmniPrism Studio — third-party, model, and runtime notices ================================================================ Release inventory ----------------- Application version: 1.0.0+1 Prepared from the model and native-worker configuration shipped with this release. Downloaded artifacts must match the SHA-256 values below before they are accepted by the application. This file covers the local AI model artifacts and native runtime components used by OmniPrism Studio. Licenses for Dart and Flutter packages are available from Settings > AI Technology > Open package licenses. The online notices page is available at: https://yyamin.com/OmniPrismStudio/ai-notices/ Model artifacts --------------- 1. BGE-M3 embedding model Artifact: BGE_M3.zip Distribution: MouseInFrance/bge-m3-int8, revision 2866e36ba13e211fa286aee1afa415d2eb1efd8e License: MIT (BGE-M3 model family; see the upstream BAAI notice) Source: https://huggingface.co/BAAI/bge-m3 Distribution source: https://huggingface.co/MouseInFrance/bge-m3-int8/tree/2866e36ba13e211fa286aee1afa415d2eb1efd8e SHA-256: 9f13b4a2d49c22c7487bcf228f699d8f49d72e729ab2e2c9deb975e0d58bbaff Required files: BGE-M3.onnx, sentencepiece.bpe.model, config.json, tokenizer.json, tokenizer_config.json, special_tokens_map.json 2. BGE reranker v2 M3 Artifact: model.onnx Distribution: kftof/bge-reranker-v2-m3-onnx-int8-avx2, revision da3c866dd7312855dbe7d114d61a4b517c5457b1 License: Apache License 2.0 (the quantized distribution states that it inherits the upstream license) Source: https://huggingface.co/BAAI/bge-reranker-v2-m3 Distribution source: https://huggingface.co/kftof/bge-reranker-v2-m3-onnx-int8-avx2/tree/da3c866dd7312855dbe7d114d61a4b517c5457b1 SHA-256: fa1f37a3b60a4994976083f04a1a4b1390fd88f4a8d2a3ee7958d443974fd00e Additional tokenizer artifact: sentencepiece.bpe.model Source revision: BAAI/bge-reranker-v2-m3, 953dc6f6f85a1b2dbfca4c34a2796e7dde08d41e SHA-256: cfc8146abe2a0488e9e2a0c56de7952f7c11ab059eca145a0a727afce0db2865 Additional tokenizer artifact: tokenizer.json Source: kftof/bge-reranker-v2-m3-onnx-int8-avx2, revision da3c866dd7312855dbe7d114d61a4b517c5457b1 SHA-256: 8bf8afbfd11306bd872018c53bfdf2e160a56f8edbcf49933324404791c148d3 3. Qwen3-VL 4B Instruct vision and query-processing model License: Apache License 2.0 Source: https://huggingface.co/Qwen/Qwen3-VL-4B-Instruct GGUF distribution source: https://huggingface.co/Qwen/Qwen3-VL-4B-Instruct-GGUF Artifact: Qwen3VL-4B-Instruct-Q4_K_M.gguf SHA-256: 66358cb18bb6b3b1b6675aa412c7a88ef01d228f481184d13668e5201c730a0a Size: 2497281664 bytes Artifact: mmproj-Qwen3VL-4B-Instruct-Q8_0.gguf SHA-256: 30ba2c7dd3127a4561b6cba9d13d0f711c91bdb38742e2f56d73c8cb596bd06d Size: 453974304 bytes The configured Qwen3-VL download URLs currently use the repository's main branch. The checksums above are the release identity; verify them again whenever the model repository or release configuration changes. 4. Qwen3 text query-parser models License: Apache License 2.0 Source family: https://huggingface.co/Qwen Qwen3-8B-Q5_K_M.gguf Source: https://huggingface.co/Qwen/Qwen3-8B-GGUF SHA-256: 068bae163faa96ad48032daf4e071a6a28fe67d8dcc95367609c2ff165e52738 Size: 5851112224 bytes Qwen3-4B-Q8_0.gguf Source: https://huggingface.co/Qwen/Qwen3-4B-GGUF SHA-256: 8c2f07f26af9747e41988551106f149b03eb9b5cb6df636027b6bf6278473300 Size: 4280404704 bytes Qwen3-1.7B-Q8_0.gguf Source: https://huggingface.co/Qwen/Qwen3-1.7B-GGUF SHA-256: 061b54daade076b5d3362dac252678d17da8c68f07560be70818cace6590cb1a Size: 1834426016 bytes The configured Qwen3 text-model download URLs currently use their repositories' main branches. The checksums above are the release identity; verify them again whenever a model repository or release configuration changes. 5. OpenAI Whisper model conversions Source artifacts: ggml-base-q5_1.bin, ggml-small.bin, ggml-large-v3-turbo-q8_0.bin Source repository: ggerganov/whisper.cpp, revision 5359861c739e955e79d9a303bcbc70fb988958b1 License: MIT Source and license: https://huggingface.co/ggerganov/whisper.cpp Upstream runtime license: https://github.com/ggml-org/whisper.cpp/blob/master/LICENSE Installed filename: ggml-base.bin (download source file: ggml-base-q5_1.bin) SHA-256: 60ed5bc3dd14eea856493d334349b405782ddcaf0028d4b5df4088345fba2efe Size: 147951465 bytes ggml-small.bin SHA-256: 1be3a9b2063867b937e64e2ec7483364a79917e157fa98c5d94b5c1fffea987b Size: 487601967 bytes ggml-large-v3-turbo-q8_0.bin SHA-256: 317eb69c11673c9de1e1f0d459b253999804ec71ac4c23c17ecf5fbe24e259a1 Size: 874188075 bytes Native runtime components ------------------------- The native versions below are the versions pinned by the worker build files. Their upstream notices and any included third-party notices remain part of the corresponding redistributable package. * ONNX Runtime 1.20.1 — MIT License https://github.com/microsoft/onnxruntime/blob/main/LICENSE The ONNX Runtime package also contains its own third-party notices: https://github.com/microsoft/onnxruntime/blob/main/ThirdPartyNotices.txt * Microsoft DirectML redistributable 1.15.2 — retain the license and notice files included in the redistributable package. https://github.com/microsoft/DirectML https://www.nuget.org/packages/Microsoft.AI.DirectML/1.15.2 * SentencePiece 0.2.1 — Apache License 2.0 https://github.com/google/sentencepiece/blob/master/LICENSE * nlohmann/json headers 3.11.3 — MIT License https://github.com/nlohmann/json/blob/develop/LICENSES/MIT.txt * whisper.cpp worker engine 1.8.3 — MIT License https://github.com/ggml-org/whisper.cpp/blob/master/LICENSE * whisper_ggml_plus local plugin — MIT License The plugin's local license is retained at: third_party/whisper_ggml_plus/LICENSE * llama.cpp-compatible runtime used by llamadart 0.8.19 — see the package and upstream notices for the exact native build included by the target platform. https://pub.dev/packages/llamadart https://github.com/ggml-org/llama.cpp/blob/master/LICENSE License text and source availability ------------------------------------- The links above identify the authoritative source and license notices. A release archive should preserve this file together with the upstream license and third-party-notice files supplied by every redistributed native package. No model or runtime component grants OmniPrism Studio permission to upload a user's library or to send document content to a remote service. Model files are downloaded only when the user requests local processing, and the application checks their configured integrity values before use.