Perplexity AI Details Ivy as HTTP Gateway

According to posts from Perplexity AI's official X account, Ivy is the HTTP gateway. It handles CPU-side request prep including parsing, tokenization, templating, and splitting large batches before sending them to Tulip over gRPC. The account states this design lets the company tune formatting and tokenization without touching inference servers. A second reply from the account describes Tulip using CUDA graphs that pre-record GPU work and LazyTensors that track results asynchronously to lower latency and boost throughput.