One connected graph
Every extraction is resolved against what we already know, so a materials advance, the lab behind it, its supply chain, and the issuers exposed to it end up on the same map.
Technology intelligence
The world's technology, as one connected graph.
tera.ai reads millions of sources and turns what they say about technology into a structured, citable knowledge graph — built for technology scouting and public-market research.
- Entities
- 5.3K
- Relationships
- 2K
- Documents
- 856
The platform
Intelligence that holds up to scrutiny
A graph is only useful for scouting and research if you can trust its edges. tera.ai is built so every one of them can be traced back to what a source actually said.
Traceable to the sentence
Every node and edge keeps the verbatim span that produced it. Nothing enters the graph without a citation you can read.
Continuous extraction
A long-running pipeline reads preprints, patents, filings, and news, and pulls out the technologies, organizations, and relationships each document actually asserts.
Built for source scale
Documents go through the Message Batches API in slices of 2,000, with several batches in flight and a cached ontology prefix, so each document costs little more than its own text.
Confidence you can act on
Edges carry a confidence score aggregated across every supporting source. Anything the fast model is unsure of is re-run on a stronger one before it lands.
Walk the connections
Traverse two or three hops out from any entity to see what links a startup's technology to its suppliers, partners, and the public companies exposed to it.
How it works
From raw documents to a citable graph
A four-stage pipeline turns what millions of sources assert about technology into connections you can search, trace, and act on.
Ingest & deduplicate
Preprints, patents, filings, and news land as documents. Each body is hashed, so a press release syndicated a thousand times is only paid for once.
Extract in batches
The worker claims unprocessed documents and submits them in batches. The model returns the entities and relationships each document asserts, with the quote behind every one.
Resolve & merge
Results are matched to existing entities by normalized name, trigram similarity, and optionally embeddings, then merged into the graph inside their own transaction.
Escalate & explore
Low-confidence extractions are re-run on a stronger model. What remains is a citable graph you can search and traverse in the explorer.
Use cases
Built for the people who scout and invest
Whether you are looking for the next capability or the issuers exposed to it, the same graph answers both questions.
Technology scouting
Find the labs and companies working on a capability before they are on anyone's list, and see who they are connected to.
- Surface who develops a capability, and who they depend on
- See partners, suppliers, and competitors in one view
- Read the source sentence behind every connection
Public-equity research
Trace a technology shift through suppliers and partners to the issuers whose economics actually move with it.
- Follow a technology through the supply chain to listed issuers
- Join the graph to market data through tickers and CIKs
- Weigh each claim by its confidence and number of sources
- Technologies
- Organizations
- People
- Markets
The knowledge graph
One map of technologies, organizations, and markets
Nodes and edges live as relational tables that scale to hundreds of millions of rows. Every edge is temporal and carries its confidence, and every claim links back to the document that made it.
- Entity types
- 10
- Relationship types
- 18
- Edges mapped
- 2K
Early and listening
What are you trying to answer?
tera.ai is early. If you scout technology or invest around it, we would like to hear what you are trying to answer.