An unsupported device should receive a clear explanation and a useful next step, not an endless loading state. Keep ordinary reading, editing, and access to saved work available wherever the product can support them independently of inference.
Define support from verified configuration and task requirements. Avoid treating a model-load attempt as the only way to discover a predictable incompatibility.
Establish capability checks and their limits
Record operating-system requirements, runtime backend, model format, available resources, and any required hardware capability. Use current runtime documentation and the exact model's official requirements as inputs.
Preflight checks can identify known incompatibility, but available memory and operating conditions can change. Handle runtime failure as well as initial eligibility and keep the diagnostic stage clear.
Review a fictional older classroom laptop
Suppose a teacher imports a PDF on a laptop that cannot run the supported local model. The app can still allow document inspection, manual lesson editing, and opening previously prepared lesson packs if those features are independent.
Test unsupported operating system, unavailable backend, insufficient resources, failed model load, and a device that becomes constrained after another app opens. Verify the user does not lose imported material or receive a false ready state.
If a smaller supported mode exists, explain what task or input limit changes. Do not imply equivalent quality without evaluation. If an optional cloud path exists, ask for an explicit choice and explain what leaves the device before using it.
Keep support messages concrete
State the unavailable capability, relevant requirement, observed condition where safe, and available alternatives. Avoid blaming the whole device when only one runtime path is unsupported.
Provide a safe diagnostic reference and version information for support. Do not require private source files to prove a hardware mismatch. Keep installation cleanup separate from user-content deletion.
The tradeoff is broad reach versus predictable support. Permitting unverified configurations can increase adoption attempts while creating confusing failures and support work. A narrower honest boundary can help users decide whether the product fits before a large download or purchase.
Revisit the support matrix after runtime, model, or operating-system changes. Compatibility can improve or regress, and a permanent hardcoded message may become inaccurate.
The ChalkLens case study documents a local desktop teaching workflow. This older-laptop scenario is fictional and does not publish a hardware requirement or promise a cloud fallback for that product.
Questions about unsupported local AI devices
Should the whole application refuse to open?
Only when its entire promised function requires the unavailable capability. Preserve independent useful work when the design supports it.
Can cloud processing happen automatically as fallback?
Do not hide that change. It affects privacy, connectivity, cost, and user expectations and needs an explicit product policy and choice.
What should a review verify first?
Force a known unsupported configuration and inspect saved-work access, explanation, and cleanup. Bring gaps to a local AI support review.