This change introduces host-specific site distiller scripts, improving how content is parsed and presented for AI-driven text distillation. We’ve refined host retrieval from page origins, improved parsing and decoding logic, and addressed unsafe buffer issues. Distiller resources have been reorganized into a dedicated directory, while unnecessary complexity and reliance on regex-based lookups have been removed.
We’ve switched from creating XML documents to HTML documents for more relaxed parsing rules, simplified the distillation process by removing events, and adopted absl::StrFormat for cleaner string handling. Namespace isolation and resource visibility controls have been tightened, and a disabled-by-default feature flag is now available for incremental rollout.
Additional improvements include better handling of profile pages on X.com (formerly Twitter), clearer metadata prefaces to help LLMs understand where the page content begins, early returns on failed script lookups, and consistency in naming and formatting. Debugging logs and other extraneous elements have been removed to streamline the final implementation.