What “Local” Means for a Home AI Assistant: Model Files and External Tasks

Introduction: How Far Does “Local” Extend?

Tom’s Hardware describes Ajax as a model designed to run on a home PC and as the always-available assistant for Odysseus, a self-hosted workspace. Google AI Edge Gallery, a separate product, says on GitHub that it runs model inference on mobile devices and can be used without an internet connection.[S2649][S2636]

When reading either product’s claims about being “local,” check where the model runs, what files it needs, which tasks use external services, and what its privacy disclosures say. Information about one product cannot establish how data flows through the other.[S2649][S2636][S2641][S115]

First Boundary: Which Device Runs the Model?

According to Tom’s Hardware, Ajax is a fine-tuned Qwen3.5-9B model designed to run on a home PC and power the Odysseus workspace. AI Edge Gallery’s GitHub description states that model inference for chat and other features runs on mobile devices.[S2649][S2636]

These descriptions identify where the models run. The provided news excerpt does not establish whether Ajax sends any data externally while carrying out tasks.[S2649]

Second Boundary: What Must Be Downloaded Before Use?

AI Edge Gallery says users can download a model from its catalog or load one they already have. A model must be available before on-device inference can begin.[S2636]

The AI Edge Gallery 1.0.19 release notes say Android gained the ability to specify auxiliary data files that must be downloaded alongside a primary model in its model allowlist. That is an Android change; it does not establish that the iPhone app has the same feature.[S2641]

When installing or changing a model, check which files it requires. Whether inference can run offline afterward and what must be downloaded during setup are separate questions.[S2636][S2641]

Third Boundary: Where Do the Assistant’s Tasks Lead?

A description of Ajax quoted by Tom’s Hardware lists search, web browsing, email, and calendar management among its everyday tasks. AI Edge Gallery’s GitHub description also describes modular Agent Skills that retrieve information from Wikipedia, maps, and URLs. These features make it useful to examine both the device running the model and the sources and tools used for each task.[S2649][S2636]

For each task you plan to use, check the product documentation and settings to see what it connects to and what information it shares. The provided material does not establish where Ajax sends data for each task.[S2649]

Fourth Boundary: What Do the Privacy Disclosures Say?

AI Edge Gallery’s iPhone App Store listing identifies IDs, diagnostics, and other data as information that may be collected and linked to a user’s identity. It lists location, usage data, and diagnostics as information that may be collected without being linked to a user’s identity. These are categories of potential collection disclosed by the developer.[S115]

The GitHub description of on-device inference and the App Store’s privacy disclosures answer different questions. The listed categories alone do not establish whether a particular prompt is transmitted or whether any particular data is actually collected.[S2636][S115]

Conclusion: Four Things to Check Beyond the Word “Local”

Even an assistant like Ajax, designed to run on a home PC, warrants a look at each connection if it performs search, email, or calendar tasks. AI Edge Gallery’s description of on-device inference, Android’s support for downloading auxiliary files, and the iPhone App Store’s privacy disclosures each address a different part of the picture.[S2649][S2636][S2641][S115]

When choosing a product, check which device runs the model, what additional files installation requires, which external tasks you will use, and the privacy disclosures for your platform.[S2649][S2636][S2641][S115]