[Expression of support]: Support from WebLLM, an in-Browser LLM inference engine #31
Description
What are the primary cross-origin storage use cases in your apps?
- AI models
- SQLite databases
- Offline storage archives
- Wasm modules
- Other
Would you use this API in production as a progressive enhancement?
Yes
Why is cross-origin storage important for your use case?
WebLLM users need to download large LLM weights and WASM modules (which include artifacts like compiled WebGPU kernels) for each AI model they intend to use. Prior to the introduction of COS, the "same-origin policy" meant that each website or extension, even if using the same model, had to cache its own copy of the model.
With careful considerations for privacy and security, the COS initiative could address this inefficiency by eliminating the overhead of downloading duplicate large models, saving both storage space (avoiding redundant caches) and time (preventing repeated downloads), and improving the experience of end users.
What company do you work for?
Concerns or reservations
No response
Additional features or improvements
No response
Any other feedback
No response