For over a decade, tracking the technology industry has meant reading an endless stream of headlines. Headlines tell you what happened today. They don't show you the structure of why it happened, or how the industry's tectonic plates are shifting beneath the surface.
A news article says "Microsoft acquires Activision Blizzard for $69 billion." That's a sentence. It can't be queried, filtered, or connected to other events. A hyperedge turns that sentence into structured data:
Every relationship has subjects (who acted), a predicate (what they did), objects (who or what was affected), a status (confirmed, rumored, denied, or failed), and the set of articles that reported on it. When multiple articles cover the same event, their evidence merges into one edge — creating a coverage-weighted measure of significance.
We call them "hyper" edges because they connect more than two nodes. A traditional graph edge links A to B. A hyperedge can link multiple subjects to multiple objects through a typed, directed relationship — capturing the actual structure of events like multi-party partnerships or regulatory actions involving several entities.
Using large language models, our system performs hyper-relational extraction on every article in the Techmeme archive. Instead of keyword matching, the extraction identifies the primary actors, the specific action taken, the target, and the context. The results are deduplicated, categorized, and linked back to source articles.
The pipeline runs every 15 minutes against new articles, and historic extraction covers the full archive back to 2024. Each edge is also embedded as a vector, enabling semantic search — you can search for "companies being forced to work with the military" and find edges about Defense Production Act invocations, even if those exact words never appear.
We've organized the graph into eight core predicates — the fundamental actions that drive the technology industry. Each page surfaces every tracked event of that type, sorted by coverage weight. Events that dominated the news rise to the top.
The knowledge graph is accessible through the TEXXR API and through several pages on this site. Semantic search uses vector embeddings to find edges by meaning, not keywords — so a query for "AI companies raising money" returns funding rounds for OpenAI, Anthropic, and Mistral even though the summaries may say "closes Series B" or "secures new investment."
GET /api/hyperedges/entity/OpenAI
GET /api/hyperedges/between/Google/Apple
POST /api/hyperedges/search
GET /api/edge/{id}/similar
Traditional news search finds articles that contain keywords. The knowledge graph finds events that involve entities — regardless of how they were described. A search for "Google antitrust" returns articles that mention those words. A query against the graph returns every regulatory action involving Google, including ones headlined "DOJ seeks breakup of search monopoly" that never mention "antitrust" at all.
The graph also reveals patterns invisible in individual articles. Which companies acquire in clusters. Which sectors face regulatory pressure in waves. How the lag between a partnership announcement and a product launch varies by industry. Whether a controversy leads to personnel changes, and how long that typically takes. These are structural questions that require structured data to answer.
The knowledge graph powers features across TEXXR. Entity profiles show every relationship involving a specific company or person. Nexus surfaces correlations and patterns between entities. Story Arc traces how topics evolve over quarters. Drift Map visualizes how entity coverage shifts semantically over time. Posts use graph data to find the constellations behind the news.
Full API documentation is available at Pulse. Topic pages are listed at Topics.