About the project
About AI Model Inspector
AI Model Inspector is a small open-source tool for looking inside model files without first installing a model runtime or uploading the model to an inspection service.
What it does
The current application reads GGUF version 2 and 3 headers, metadata, and tensor descriptors. It also reads selected ONNX protobuf fields, including model properties, graph inputs and outputs, nodes, initializers, tensor types, and top-level graph relationships.
From that information it can estimate model-weight and KV-cache memory, project size ranges for common GGUF quantization profiles, compare two files of the same format, export parsed results as JSON, and calculate a SHA-256 fingerprint on demand.
Why local inspection is useful
Model files can be large, private, or inconvenient to move. The parsers and hash worker run in the browser, and AI Model Inspector has no model-upload endpoint. Local processing also makes it easier to inspect a file directly from removable or local storage. The tradeoff is that parsing uses your device's memory and CPU, and a very large or malformed file can be slow.
Open source
The source is available in the AI Model Inspector GitHub repository. The site is a static Vite application with no application backend. Format-specific parsing is implemented in JavaScript so its behavior can be reviewed alongside the documentation.
If the tool misreads a file or an estimate looks wrong, open an issue with enough non-sensitive detail to reproduce the problem. Code changes and documentation corrections can be proposed through the repository.
Start inspecting
Open the inspector to choose a GGUF or ONNX file. If you want to understand a field before loading a model, begin with the technical documentation.