Pipit
An open-source offline AI universal search engine for Android. Built on on-device multimodal models and OCR, it provides semantic search across domains including photos, PDFs, messages and contacts.
Timeline (last 90 days)
Developer sanjay434343 released an open-source project called Pipit on DEV Community, positioned as a fully open-source Android AI universal search engine that runs 100% on-device.
Pipit uses on-device Google ML Kit to perform OCR text recognition on PDFs, receipts and screenshots, searching the text inside files rather than just file names.
Pipit supports retrieving photos from the gallery using natural-language descriptions, relying on a quantized multimodal model deployed on the phone to understand image content rather than traditional tags or metadata matching.
Pipit integrates seven domains—photos, PDFs, messages, contacts, videos, audio and apps—into a single search entry point.
All of Pipit's processing is done on the phone's chip via a quantized Google LiteRT multimodal model, and the app can run fully air-gapped, emphasizing zero cloud and zero tracking.
Pipit claims acceleration via local Vulkan GPU/NPU delegation, with search latency kept under 50 milliseconds.
The actual performance of the 50ms latency claim depends heavily on whether the device supports the corresponding delegation backend, whether the model operators are fully covered, and the size of the index.
Pipit's source code and release APKs are both hosted on GitHub (github.com/sanjay434343/Pipit).
Pipit relies heavily on Google's ML Kit and LiteRT ecosystem, and its model size and hardware acceleration capabilities are limited by the support scope of Google's toolchain.
The quality of Pipit's semantic visual search depends heavily on the precision of the quantized model, and recall may drop on low-end devices where the compression rate is higher.
1 more timeline events
All Facts (11)
The time and battery cost of Pipit's first full scan across seven domains to build the index are not disclosed in the original text.
50%UnverifiedPipit uses on-device Google ML Kit to perform OCR text recognition on PDFs, receipts and screenshots, searching the text inside files rather than just file names.
50%UnverifiedPipit supports retrieving photos from the gallery using natural-language descriptions, relying on a quantized multimodal model deployed on the phone to understand image content rather than traditional tags or metadata matching.
50%UnverifiedPipit integrates seven domains—photos, PDFs, messages, contacts, videos, audio and apps—into a single search entry point.
50%UnverifiedAll of Pipit's processing is done on the phone's chip via a quantized Google LiteRT multimodal model, and the app can run fully air-gapped, emphasizing zero cloud and zero tracking.
50%UnverifiedPipit claims acceleration via local Vulkan GPU/NPU delegation, with search latency kept under 50 milliseconds.
50%UnverifiedThe actual performance of the 50ms latency claim depends heavily on whether the device supports the corresponding delegation backend, whether the model operators are fully covered, and the size of the index.
50%UnverifiedPipit's source code and release APKs are both hosted on GitHub (github.com/sanjay434343/Pipit).
50%UnverifiedPipit relies heavily on Google's ML Kit and LiteRT ecosystem, and its model size and hardware acceleration capabilities are limited by the support scope of Google's toolchain.
50%UnverifiedThe quality of Pipit's semantic visual search depends heavily on the precision of the quantized model, and recall may drop on low-end devices where the compression rate is higher.
50%UnverifiedDeveloper sanjay434343 released an open-source project called Pipit on DEV Community, positioned as a fully open-source Android AI universal search engine that runs 100% on-device.
50%