DigUp
An open-source Mac desktop application that implements semantic search using locally running multimodal embedding models. It supports natural-language retrieval across text, images, audio, and video, operates entirely offline, and requires no internet connection.
Timeline (last 90 days)
DigUp is an open-source Mac semantic search application based on the EmbeddingGemma 2 model that runs entirely locally.
DigUp can map text, images, audio, and video into the same vector space, enabling cross-media semantic retrieval.
DigUp splits audio and video into 30-second windows, takes one frame per shot, and then uniformly encodes them into the vector space.
DigUp uses the Q8_0 GGUF quantized model provided by ggml-org, approximately 865 MB, which only needs to be downloaded once.
DigUp runs on llama.cpp with Metal acceleration, wrapped in a native Swift application, with no dependency on a Python environment.
During search, DigUp loads only approximately 250 MB of the text encoder and returns results approximately 0.1 seconds after the user stops typing.
DigUp keeps peak memory usage during indexing below 2 GB, and the auxiliary process exits automatically after indexing is complete.
DigUp only connects to the network when downloading the model and checking for updates, and update checks can be manually disabled.
DigUp is free and open-source under the MIT license, signed and notarized, and targets Apple Silicon chips and macOS 14 or later.
The project repository for DigUp is located at GitHub ARahim3/DigUp.
1 more timeline events
All Facts (10)
DigUp is an open-source Mac semantic search application based on the EmbeddingGemma 2 model that runs entirely locally.
50%UnverifiedDigUp can map text, images, audio, and video into the same vector space, enabling cross-media semantic retrieval.
50%UnverifiedDigUp splits audio and video into 30-second windows, takes one frame per shot, and then uniformly encodes them into the vector space.
50%UnverifiedDigUp uses the Q8_0 GGUF quantized model provided by ggml-org, approximately 865 MB, which only needs to be downloaded once.
50%UnverifiedDigUp runs on llama.cpp with Metal acceleration, wrapped in a native Swift application, with no dependency on a Python environment.
50%UnverifiedDuring search, DigUp loads only approximately 250 MB of the text encoder and returns results approximately 0.1 seconds after the user stops typing.
50%UnverifiedDigUp keeps peak memory usage during indexing below 2 GB, and the auxiliary process exits automatically after indexing is complete.
50%UnverifiedDigUp only connects to the network when downloading the model and checking for updates, and update checks can be manually disabled.
50%UnverifiedDigUp is free and open-source under the MIT license, signed and notarized, and targets Apple Silicon chips and macOS 14 or later.
50%UnverifiedThe project repository for DigUp is located at GitHub ARahim3/DigUp.
50%