One question in. A sourced brief out.
The Master Disease Agent, the research dashboard and the REST API all run on the same engine. Use whichever fits how you work.
Find Your Cure
Ask about a disease in plain language and get an outlook, ranked treatments, safety flags and historical approval odds.
For researchersResearch dashboard
The full brief: every compound, trial, protein target and paper the sources returned, plus rating and feedback.
For developersREST API
Every capability as a documented endpoint, from the full brief to individual source lookups.
One query, four databases, one checked ranking.
This simulation shows what happens in the minute after you name a disease: parallel retrieval, AI ranking, then a rules-based safety check that moves withdrawn drugs down whatever the AI said.
What every analysis contains.
Run an analysis on any disease name. A brief usually takes under a minute, depending on how quickly the upstream databases respond.
Ranked treatment candidates
Clinical stage, mechanism, primary target, evidence grade (A/B/C), AI confidence and historical approval odds, each linked to its sources.
Safety flags
Drugs ChEMBL lists as withdrawn, and drugs whose trials were terminated, withdrawn, suspended or never reported, are flagged and ranked down.
Clinical trial landscape
Registered trials with NCT IDs, phase, recruitment status, interventions and stop reasons.
Biology
Reviewed protein and gene targets from UniProt, with function summaries and drugs known to act on each.
Summary, pathways and gaps
A cross-source summary, the key biological pathways, and what the evidence doesn't answer yet.
Transparency
Which sources answered, which failed, how long each took, and an automated grounding score.
Source-level tools.
Each connector is also available on its own and returns structured records straight from the source.
| Endpoint | Source | Returns |
|---|---|---|
| GET /api/v1/intelligence/analyze?disease= | All four + AI | Full brief with candidates, flags and approval odds |
| GET /api/v1/intelligence/trials?condition= | ClinicalTrials.gov | Trials with a phase and status summary |
| GET /api/v1/intelligence/compounds?disease= | ChEMBL | Compounds by indication and phase |
| GET /api/v1/intelligence/targets?disease= | UniProt | Reviewed protein targets |
| GET /api/v1/intelligence/targets/gene/{symbol} | UniProt | One protein by gene symbol |
| GET /api/v1/encyclopedia/diseases?q= | Disease catalog | Diseases by name or synonym |
| POST /api/v1/publications/ingest | PubMed | Background literature ingestion job |
| GET /api/v1/status | All | Live health check of every connector |
Full, interactive reference: /docs.
It gets better with use.
Every brief is stored. When someone analyzes a related disease later, relevant past runs and researcher notes are recalled as hints. Rate a brief and that rating shapes what gets recalled next time. Every candidate must still come from freshly retrieved data.
Serverless, end to end.
Python FastAPI on Vercel Functions, async connectors queried in parallel with per-source timeouts, an OpenAI model for structured synthesis, and Postgres (Neon) for runs, memory and the disease catalog. Connector health is public on our status page.