For researchers using AI tools
AI Agents
The Atlas is machine-readable on purpose. Every record is served as structured JSON, every page carries schema.org markup, and the relationships between records are explicit. These prompts point an AI assistant at that data instead of letting it improvise.
Copy-paste prompts
Each one is wired to this site's live API. Paste into Claude, ChatGPT, Gemini or any assistant that can fetch a URL.
Using only https://crispr2.2.25.209.181.nip.io/api/v1/entities?kind=treatment as your source, list every gene-editing treatment whose status is "Approved". For each, give the disease it treats, the technology it uses, the company, and the URL of its Atlas page. If a field is missing in the data, say "not recorded" rather than filling it in from memory. Finish by stating plainly that approval in one country does not mean availability everywhere.
Fetch https://crispr2.2.25.209.181.nip.io/api/v1/entity/disease/DISEASE-SLUG (replace DISEASE-SLUG, e.g. sickle-cell-disease). Write a 300-word briefing for an intelligent non-specialist covering: what the disease is, which genes are involved, what gene-editing approaches are being tried, which companies and trials appear in the related records, and — most importantly — the evidence level attached to each claim. Do not present anything preclinical as if it were an available treatment.
Fetch https://crispr2.2.25.209.181.nip.io/api/v1/trials and group the results by phase. For each phase, list the trials with their disease, sponsor and NCT identifier. Flag any trial whose status is Paused, Terminated or Discontinued and say so prominently. Note that Atlas records carry a last-updated date and that ClinicalTrials.gov is authoritative for current status.
Fetch https://crispr2.2.25.209.181.nip.io/api/v1/entities?kind=technology. Build a comparison table of CRISPR-Cas9, base editing and prime editing covering: what it edits, whether it cuts both DNA strands, precision, cargo size, delivery difficulty and clinical maturity — using only the fields in the data. Then write two paragraphs explaining, to a 15-year-old, why not cutting both strands matters.
Fetch https://crispr2.2.25.209.181.nip.io/api/v1/entity/company/COMPANY-SLUG (e.g. crispr-therapeutics). Summarise the company's disclosed pipeline stage by stage, name its partners and the diseases it targets, and list its related trials. State the record's last-updated date at the top. Add: this is educational information, not investment advice, and pipelines change frequently.
Fetch https://crispr2.2.25.209.181.nip.io/api/v1/entity/technology/TECHNOLOGY-SLUG (e.g. base-editing) and use the "simple" field as your starting point. Explain the technology to a curious 11-year-old in under 200 words, using one analogy — then add one short paragraph explaining exactly where that analogy stops being accurate. Keep it honest rather than exciting.
The API
Free, read-only, no key. Please cache and link back.
/api/v1/entities?kind=…
Every record of a kind: technology, disease, treatment, trial, company, scientist, gene, paper, institution, ethics, agriculture, learn, glossary.
/api/v1/entity/{kind}/{slug}
One record in full, including every related record in the knowledge graph.
/api/v1/trials
The trial database, filterable by phase and status.
/api/v1/news
Headlines, publishers and links. Article text stays with its publisher.
Why the structure matters
Entities, not articles
A disease, a gene, a trial and a company are separate records with stable URLs — so an assistant can follow a relationship instead of guessing at one.
Evidence is a field
Every record carries an explicit evidence level, so a model can tell an approved medicine from a mouse experiment without reading between the lines.
Answers near the top
Each page opens with a direct one-sentence answer before the detail, which is what both readers and retrieval systems actually need.