Fiduciary AI for clinical practice

Practice management plus clinical evidence. Designed by physicians, for physicians.

Two agents that fit the way you actually work. Arohi runs your practice intelligence. Asha answers clinical questions with citations you can verify. Both are accessible through one API.

Try Arohi via API Schedule a clinic demo

Built by physicians · HIPAA-conservative · Vertex AI BAA backbone · Audit trail on every answer

Two products for clinicians

Different jobs, same fiduciary architecture, same API surface.

🏥

Arohi: practice management intelligence

Arohi (Sanskrit for ascending) is your practice-side AI. She manages patient longitudinal records, scheduling, billing context, and clinical alerts. She speaks clinical terminology (the user is a physician) and never interacts with patients directly.

Read the Arohi deep dive →

Asha (clinical mode): evidence-grounded medical answers

Asha is our biomedical fiduciary agent. The consumer chat lives at askasha.org. For clinicians, access her through A2A with agent_id: asha. She returns the evidence behind every answer (PMIDs, StatPearls IDs, FDA label citations) so you can verify before you act.

Read the Asha deep dive →

Why doctors choose DNAi

We built this because the existing options either hallucinate or refuse to engage. Both are unacceptable in clinical work.

🩺

Built by practicing physicians

Deepan Singh, MD, FAPA (double board-certified psychiatrist) and Paridhi Anand, MD (double board-certified pediatrician) co-founded DNAi Systems. Every architectural decision was reviewed by clinicians before it shipped.

🔍

Audit trail on every answer

Every response logs the retrieved evidence behind it. If a patient or colleague asks "where did that come from?", you can answer with the underlying citations. No black box.

🛡

Sacred refusals are architectural

Asha will not prescribe medications, give a definitive diagnosis, fabricate citations, or override an emergency signal. These are encoded constraints, not policies that could be talked around.

📚

Citation precision

When evidence is thin, Asha says so. When two guidelines disagree, she shows both and labels the strength of each. Honest uncertainty is part of clinical reasoning, and the system reflects that.

🌍

Multilingual at the point of care

Eleven languages including Hindi, Spanish, Mandarin, and Arabic. Patient-language outputs are available when you need them, with the same evidence base.

📱

API-first integration

No new chair to sit in. Hit our A2A endpoints from your EHR, your custom workflow, or your patient portal. Authentication is a single bearer key.

The evidence corpus behind Asha

Every clinical answer is grounded in retrieved evidence. These are the indexed sources.

5.1M
PubMed Abstracts
16.5M
OpenAlex Top-Cited
5.2M
PMC Full Text
889K
DailyMed Drug Labels
76K
StatPearls Articles
39K
FDA Drug Labels
206K
Medical Codes
829
Total Collections

Sources: indexed Qdrant collections (829 collections, ~126M+ vectors as audited 2026-05-28). Includes clinical guidelines (AAP, ACOG, AAOS, AGS, NCI PDQ), DailyMed, FDA labels (v1 + v2), OpenFDA adverse events, drug interaction knowledge, LactMed, rare-disease references (Orphanet, GeneReviews, HPO), and specialty guideline stacks (psychiatry, neurology, oncology, emergency, geriatrics, pediatrics, humanitarian medicine).

Compliance posture

HIPAA conservatism is the default. Concrete and documented.

Vertex AI BAA backbone

Cognition runs on Google Vertex AI under a Business Associate Agreement. PHI handling is HIPAA-aligned by infrastructure choice.

TLS 1.3 everywhere

All API traffic is encrypted in transit. No plaintext PHI on the wire, ever.

PHI redacted before logging

Per the draft API Terms, PHI is redacted before storage in audit logs. Queries are still logged for security and abuse prevention.

No PHI on marketing surfaces

The Asha consumer chat at askasha.org carries no marketing pixels. The patient bridge on this site is Plausible-only. PHI never touches a third-party tag.

No training on customer data

API inputs and outputs are not used to train any of our models or any underlying vendor models. Use is for inference only.

Audit log retention

Per-query audit log (caller, agent, tokens, latency, IP) is maintained for security, abuse prevention, and incident response. Available on request for your covered entity.

Bring fiduciary AI into your practice

Start with a free API key on the developer playground. Or email us about a clinic-specific demo and integration discussion.

Try the API

Or email [email protected] for a clinic demo