ShipSafe Filter
Most model license trouble starts with one clause nobody checked: a user-count cap, an attribution duty, a non-commercial line. Set the constraints your product actually has below, and the Filter shows which of the 200 open models in the LocalAITools dataset clear all of them, and which miss by exactly one.
Loading the LocalAITools dataset...
Near misses: out by exactly one constraint
These models fail a single one of your active constraints. The failing constraint is named with the dataset's recorded terms, so you can judge whether it is one you can live with.
How to read this: the filter applies the license terms recorded in the LocalAITools dataset mechanically, field by field. It is not legal advice, and it is not a lawyer's reading of the full license text. The human-checked version of this check is the site's paid license clearance report.
How each constraint is read
- Commercial use is the dataset's own yes or no field, recorded per model. Eight of the 200 models are recorded as not allowing commercial use.
- Closed product passes when the dataset allows redistribution and modification and the recorded terms carry no share-alike duty. Passing use restrictions downstream (the RAIL, DeepSeek and Gemma pattern) is not counted as copyleft: it binds how the model is used, not whether your product stays closed.
- Attribution fails when the recorded restriction text requires attribution, a notice file, branding such as "Built with Llama", displaying the model name in the product, or retaining copyright and license notices. Where the dataset records no restriction text, no attribution duty is recorded either.
- User-count caps are counted only when the recorded text ties a monthly-active-user threshold to a separate license, permission or authorization: the Llama pattern (700M) and the older Qwen pattern (100M). A threshold that only triggers a branding display, as in the Kimi K2 terms, is not a cap, and revenue thresholds are not user-count caps.
- Expected monthly users uses the recorded cap exactly: a model is excluded only when its cap is below the number you enter. Sitting exactly at the cap still clears, because the recorded terms trigger above the threshold.
Where the data comes from
The Filter is generated from the same model-license dataset that powers LocalAITools: ... open models across ... licenses, each with recorded commercial-use, redistribution and modification fields and, for ... models, a restriction note. A few restriction notes arrive shortened from the dataset feed and end with an ellipsis; the Filter reads the text exactly as recorded, and every result row shows the note so you can judge it yourself. Terms change, so confirm the current license on the model's own page before you rely on a result.