Perplexity lets AI write search code in Python

Perplexity's Search as Code lets AI models write custom Python scripts to run search workflows instead of using fixed APIs. The scripts run in a secure sandbox and use the Agentic Search SDK to filter, deduplicate, and rerank results. Models define search strategies directly, enabling parallel queries and programmatic filtering. This reduces token usage and improves result precision by avoiding junk in context windows. The architecture separates search logic from the model, allowing deeper, more precise research workflows.