Research Hub

Three research workflows in one interface, all accepting natural language queries and returning synthesised responses with original sources cited and attached. Results stream via Server-Sent Events in real time — the user sees search progress, scraping status, and synthesis output as it happens rather than waiting for a final result.

INVESTIGATE mode runs entity-focused research: named entity recognition extracts people, places, and organisations from the query, geocoding validates and maps location mentions, web scraping pulls full content from relevant sources, and the results are synthesised into a structured response with every claim linked to its source document and URL.

DEEP mode generates lateral queries from the original question — exploring adjacent topics and alternative framings the user might not have considered. Results are clustered by theme, analysed for entity overlap and topical patterns, and presented with the lateral connections highlighted. Designed for research where the most useful answer might not come from the most obvious search.

RECALL mode grounds queries exclusively against internal knowledge — conversation archives, session logs, scraped content, code repositories, and indexed documents. No external search. Returns what the system already knows about a topic, with citations back to the original internal sources. Useful for answering 'have we seen this before' and 'what do we already know about X' without hitting the web.