The research surface for people who still cite primary sources.

Every SEC filing a company has ever made is indexed and XBRL-parsed, so three periods of income statement can be normalised and compared without anyone touching Excel. 8-Ks are laid out as an event timeline with FinBERT sentiment scoring — you see inflection points, not a stack of PDFs.

Pro Plus accounts get semantic search across the entire filing corpus, powered by Voyage embeddings. "Find every time this management team described margin pressure in the last five years" becomes a query, not a project. Risk-factor diffs flag additions, deletions, and tone shifts year-over-year — the AI annotates the change, not the paragraph.

An equity thesis rarely lives in the income statement alone. Convexity wires Form 4 insider flow, 13F holdings, congressional trades, USPTO patents, H-1B talent flows, ATS job postings, and transcript search into the same ticker view. The alt-data sits beside the financials rather than inside a separate subscription.

Insider signal intelligence reads transactions with context — cluster detection, officer-role weighting, transaction-type classification. A CEO open-market purchase doesn't look like a 10b5-1 plan sale. FedSpeak NLP layers FOMC speeches, minutes, and FRED series on the same surface, useful for the days your sector thesis is secretly a rates thesis.

Convexity's AI stack is routed by tier: Claude Haiku on Free, Sonnet on Pro, Opus on Pro Plus, with an OpenAI fallback for uptime. Retrieval runs on Voyage embeddings over the filing corpus, so the model answers from primary sources rather than guessing. Every claim ships with a citation; if the citation isn't there, the claim isn't made.

Pro Plus also includes 20 deep-research runs per month — long-running, multi-source synthesis jobs for the weeks when "read everything on this name" is the actual task. A three-paragraph risk-factor comparison cross-checked against a two-week-old insider cluster becomes one prompt instead of a forty-minute session on EDGAR.