OIC ("Oh, I See") is a local visual-watching project. The current iOS app lives in ItIsToasted/ and is being extended from toast monitoring toward promptable on-device watchers such as cat-door monitoring.
ItIsToasted/: iOS app workspace and sourceRuns/: captured run artifacts and experiment output
The iOS app currently depends on a locally installed third-party runtime that is not checked into git:
- required path:
ItIsToasted/ItIsToasted/ThirdParty/llama.xcframework
Before building the app, download the official llama.cpp iOS XCFramework release and place llama.xcframework at that path.
Current runtime used by this repo:
- runtime:
llama.cppiOS XCFramework - local install path:
ItIsToasted/ItIsToasted/ThirdParty/llama.xcframework - source release family: official
ggml-org/llama.cppXCFramework release artifact
The app no longer depends on CocoaPods Gemma inference packages. The active local model path is the vendored llama.cpp XCFramework plus local GGUF files.
The Gemma GGUF model files are local-only and are intentionally excluded from git.
Current model defaults used by the app:
- primary model filename:
google_gemma-4-E2B-it-Q4_K_S.gguf - projection model filename:
mmproj-google_gemma-4-E2B-it-f16.gguf
Canonical runtime location:
- the app Documents folder under
Models/ - the app expects the files in:
Documents/Models/google_gemma-4-E2B-it-Q4_K_S.ggufDocuments/Models/mmproj-google_gemma-4-E2B-it-f16.gguf
Current development/simulator source files on this machine:
/Users/zhian/Downloads/bartowski/google_gemma-4-E2B-it-GGUF/google_gemma-4-E2B-it-Q4_K_S.gguf/Users/zhian/Downloads/bartowski/google_gemma-4-E2B-it-GGUF/mmproj-google_gemma-4-E2B-it-f16.gguf
Important behavior:
- the iPhone app cannot read your Mac download folder directly
- do not keep the
.gguffiles underItIsToasted/ItIsToasted/Models/, because Xcode will bundle them into the app and make the build enormous - instead, copy the files into the app’s Documents area as a separate operation
- if Finder only lets you drop files onto the app root, that is okay: the app will move the expected
.gguffiles from the Documents root intoDocuments/Models/when you useRefresh Gemma statusorTest Gemma runtime - the app is currently coded to use these exact filenames
- Install CocoaPods dependencies for
ItIsToasted/. - Download the official
llama.cppiOS XCFramework and place it atItIsToasted/ItIsToasted/ThirdParty/llama.xcframework. - Open
ItIsToasted/ItIsToasted.xcworkspace. - Build and run the app once so its Documents container exists.
- Use Finder/iPhone file sharing to copy the two GGUF files into the app.
- If Finder only accepts drops on the app root, drop them there.
- Launch the app and use
Refresh Gemma statusorTest Gemma runtime; the app will move the files intoDocuments/Models/before checking them.
The cat-door watcher currently uses the GGUF Gemma runtime for startup validation and frame capture tracing only.
- it does not currently perform true image-conditioned Gemma inference on captured door frames
- the earlier first-frame multimodal experimentation was removed because it did not match the GGUF assets used by this repo
- the next valid multimodal step must use a real GGUF-compatible iOS API for image ingestion