Comment by drodgers

It should just accept a docker container (which normal lambdas can) or a WASM blob that exposes HTTP over a standard port.

Could use optional headers for extra integration points (eg. better per-request resource usage tracking for scaling).