Show HN: Airgap-check – find what your AI app silently downloads offline (github.com)

A new tool named airgap-check has been introduced to help developers identify what their AI applications attempt to download when operating offline or in isolated environments. In many AI frameworks, such as those utilizing large language models (LLMs), it is often assumed that pointing to a local instance will suffice for offline functionality. However, certain libraries may still attempt to fetch resources like model weights or tokenizers on first use, leading to potential failures or long delays when deployed in air-gapped networks. Airgap-check addresses this issue by performing static scans of code repositories to detect calls that may attempt to reach external servers and by running applications in a Docker container with no network access to log failed connection attempts. This tool is significant for the AI/ML community, especially in sectors like healthcare and defense, where strict security protocols require applications to operate without internet access. Airgap-check enhances the ability to preemptively identify and mitigate potential network dependency issues by displaying clear insights into required resources and providing ready-to-implement remediation suggestions in the form of Dockerfile snippets. Through its findings, airgap-check fosters a more robust approach to developing secure AI applications, ensuring that developers are aware of the offline capabilities of their tools before final deployment.