Research access

Bulk and programmatic access to the Robertium hypothesis catalog and knowledge graph

For most researchers, the public web catalog at /hypotheses is enough. If you need bulk download, programmatic queries, or a copy of the underlying graph for an offline analysis, email and we'll work it out. There is no signup or paywall — this page describes what is currently available and what is planned.

Available now

  • Hypothesis catalog as JSON — the same file the website renders is publicly served at /data/hypotheses.json. No authentication. Includes outreach-quality hypotheses with full evidence chains (PMIDs, titles, scores, mediator/drug/outcome metadata). Updated when the cross-domain orchestrator is re-run.
  • OpenAPI 3.1 specification — full schema for every public JSON endpoint (catalog, prospective tracking, provenance bundles, trials indexes). Browse the interactive reference or grab the raw /openapi.yaml to import into Postman, Insomnia, or Swagger Editor.
  • Knowledge graph dump (Kuzu format) — available on request. The complete graph with all 160,955 entities and 270,791 claim edges across ten therapeutic domains. Useful if you want to run your own ABC queries, swap in different filters, or merge with other graphs.
  • Raw claims database — available on request. Postgres dump of the structured claims table (subject, predicate, object, polarity, confidence, source PMID) for all 90,166 papers that passed the L1 biomedical-relevance filter. Useful if you want to apply your own extraction quality thresholds or re-derive hypotheses.
  • Open-source pipeline — the full ingest, extraction, and orchestration code is at github.com/routewise96/robertium under MIT. Clone the repo, drop in an OpenAlex concept ID for a new domain, and run the pipeline on your own corpus. Cost on cloud LLM APIs is roughly $3–$7 per 5,000 papers.

Planned

A REST API for direct programmatic queries against the live catalog and graph. The intent is to expose endpoints along these lines:

  • GET /hypotheses?drug=X&mediator=Y&outcome=Z — filter the catalog by any combination of fields
  • GET /hypotheses/{id} — single hypothesis with full evidence chain
  • GET /entities/{name} — all claims and cross-domain occurrences for a named entity
  • GET /evidence/{pmid} — every claim extracted from a given paper

What ships first depends on what people actually need. If you have a use case — even a rough one — email it. That signal is more useful than a roadmap.

Request access

For the graph dump, the raw claims database, or to discuss a programmatic use case:

Email daniel@robertium.com Clone the pipeline