Summary
Catalog installed state is not reconciled with local inventory; completed downloads do not establish a selected canvas model. The default Use Cloud handler only logs to the console. Engine targeting is limited, and there is no LLM category tab despite LLM entries. hardware_evaluator.py:83 evaluates total rather than available VRAM and shows fixed speed bands such as <1s without workload/device benchmarking.
Environment and evidence
- Review finding: F30 (2026-10-05 product/technical review).
- Baseline: Windows,
dev at c1c5dbe. The remote dev matched this commit when filing.
- Evidence: Code + Browser.
- Priority recommendation: P2 - material improvement or integration validation. This is review triage, not a production-incident severity declaration.
- No real credentials, paid inference, engine installation, or unrelated process termination were used for review probes. Mocked observations establish the stated code behavior, not live-provider/GPU acceptance.
Reproduction or validation
Inspect ModelHubView and catalog/inventory integration: default Use Cloud only logs, install state is not reconciled into catalog state, and downloads do not establish a selected canvas model. LLM entries have no category tab. Hardware evaluation uses total VRAM and fixed speed bands despite receiving free VRAM. No latency benchmark was performed.
User impact
Discovery, installation, readiness, and use feel like disconnected screens; performance promises may be misleading.
Proposed approach
Refresh inventory after download, show actual installed/readiness state, support Use Model/Use Provider with validated capabilities, and label estimates with their assumptions.
Acceptance criteria
Additional source references
Verification scope
Real GPU inference, paid-provider compatibility, and a clean-machine packaged desktop journey remain unverified. Any follow-up implementation should target dev under the repository's contribution/branching rules.
Summary
Catalog installed state is not reconciled with local inventory; completed downloads do not establish a selected canvas model. The default Use Cloud handler only logs to the console. Engine targeting is limited, and there is no LLM category tab despite LLM entries. hardware_evaluator.py:83 evaluates total rather than available VRAM and shows fixed speed bands such as
<1swithout workload/device benchmarking.Environment and evidence
devat c1c5dbe. The remotedevmatched this commit when filing.Reproduction or validation
Inspect ModelHubView and catalog/inventory integration: default Use Cloud only logs, install state is not reconciled into catalog state, and downloads do not establish a selected canvas model. LLM entries have no category tab. Hardware evaluation uses total VRAM and fixed speed bands despite receiving free VRAM. No latency benchmark was performed.
User impact
Discovery, installation, readiness, and use feel like disconnected screens; performance promises may be misleading.
Proposed approach
Refresh inventory after download, show actual installed/readiness state, support Use Model/Use Provider with validated capabilities, and label estimates with their assumptions.
Acceptance criteria
Additional source references
Verification scope
Real GPU inference, paid-provider compatibility, and a clean-machine packaged desktop journey remain unverified. Any follow-up implementation should target
devunder the repository's contribution/branching rules.