To ensure content provenance, we should integrate the C2PA specification into our pipelines.

After the AI subnet is stable we should add support for more custom containers.

Provide gateways with a way to see which parameters were used during inference.

For long-running requests, providing users with intermediate progress updates would be beneficial.

To ensure network quality on the AI subnet, we should implement job result verification.

INC asked us to provide the gateway with GPU stats of the runner containers. This data can be used by their aggregator.

We should create a generic inference job type between pipelines.

We should cleanup the started containers and release VRAM when the worker is shut down.

Implement O+W split like with transcoding.

To enable more GPUs to join the network we should allow GPUs to run the same capability.

To improve the developer experience it is important to create client SDKs.

We should implemented fine-tuned models to allow Os to serve a bigger range of models.