Day 101: Licenses: what you can actually ship with
The boring detail that can sink a product
'Open' doesn't mean 'do anything'. Model licenses vary wildly — some are truly permissive (Apache 2.0, MIT), some have use restrictions or scale caps (Llama's community license limits very large deployments and has acceptable-use terms), some restrict commercial use. Since FitXpert is a commercial product with Schemax implications, you must confirm each model's license permits your intended use *before* building on it.
- Apache 2.0 / MIT — permissive, commercial-friendly (many Qwen, Mistral, Phi releases).
- Llama Community License — commercial use allowed but with an acceptable-use policy and a scale threshold that triggers extra terms.
- Non-commercial / research-only — some releases forbid commercial use entirely; unusable for a product.
- Fine-tune output terms — check whether a model restricts using its outputs to train other models (relevant to Stage 4).
Read the license, don't assume
This is unglamorous and genuinely important — building a product on a non-commercial model is a discovery you want to make on Day 101, not after launch. It's also an interview signal: an engineer who checks licenses understands that shipping AI is a legal and business act, not just a technical one.
Key terms
- Model license
- The legal terms governing how a model's weights and outputs may be used — permissive, restricted, or non-commercial.
- Acceptable-use policy
- License terms restricting certain uses of a model even when commercial use is broadly permitted.
Why must you check a model's license before building FitXpert on it?
