Asha AI API

Asha, in your product.

Evidence-grounded medical answers over one HTTPS call, with the sources attached to every response. Built for teams building healthcare AI. Keys from this page reach Asha only.

Read the docs
  • Sources and citations on every answer
  • OpenAI-compatible, A2A v1.0, one-file CLI, MCP server
  • Asha-only scope, PHI redacted before storage, self-serve BAA

Get an API key

250 queries a month, 50 a day (20 RPM). No card. Extended 1,000-query evaluation tier available through January 2027.

Build

Call Asha the way you already work

An HTTPS call, the OpenAI SDK you already use, a one-file CLI, or a tool inside your coding agent.

One POST. The answer comes back with its sources. Create a key above and it fills in here.

curl · POST /a2a/v1/message:send
curl -X POST https://api.askasha.org/a2a/v1/message:send \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "agent_id": "asha",
    "message": {
      "role": "user",
      "parts": [{"text": "Which antihypertensives are safe in pregnancy?"}]
    },
    "metadata": {"audience": "clinician", "session_id": "visit-2041"}
  }'

Already on OpenAI SDKs, LangChain, LiteLLM, or Vercel AI? Change base_url and the key: Drop-in with OpenAI SDKs. Terminal: Command line. Every field: Send a message and Stream a response.

White-label itYour product's name, the reader, the length, emoji or none. Safety rules and citations stay on.
Customize
Use cases

What teams build with Asha

One API, tuned per request for the person reading the answer. Every example copies as a ready request body.

Patient apps and portals

Plain-language answers with sources inside your app or patient portal.

What does an A1c of 7.8 mean, and what should I ask my doctor?
audience: patient
Settings reference

Caregiver support

Practical guidance for family members caring for someone at home.

My mother has dementia and has stopped eating much. What should I watch for?
audience: caregiver
Settings reference

Telehealth intake and visit prep

Short, focused answers that help patients arrive ready for the visit.

Help me prepare questions for my follow-up after a knee replacement.
audience: patientverbosity: brief
Settings reference

Clinician decision support

Peer-level answers with citations for physicians at the point of care.

First-line outpatient treatment for community-acquired pneumonia in an adult with no comorbidities?
audience: physicianverbosity: detailed
Settings reference

Pharmacy and medication questions

Interactions, dosing context, and label facts drawn from DailyMed and FDA labels.

Does ibuprofen interact with lisinopril?
audience: pharmacist
Settings reference

Nurse lines and triage support

Clinical answers for nurses. Emergency routing stays on under every setting.

Caller reports chest pain spreading to the left arm. What are the next steps?
audience: nurse
Settings reference
Benchmarks

Measured against the best

Every run is reproducible from the live benchmark register at dnai.systems/benchmarks and GET /api/benchmarks.

Clinical AI Systems

Vishwanath et al. 500-Item 3-Judge Panel (Nature Medicine 2026 rubric)

n = 500
Asha Apex-J63.8%
Frozen 20 Sep 2026; +8.9 points over bare GPT-6 Astra (sign test p = 2.0e-16)
OpenEvidence62.6%
Published clinical reference system
UpToDate Expert AI61.3%
Published clinical decision support
GPT-6 Astra (Bare)54.9%
Unassisted frontier baseline

Frozen Apex-J, the served Apex tier, under the study's own three-judge panel: +8.9 points over bare GPT-6 Astra on the same 500 answers (sign test p = 2.0e-16).

HealthBench Hard

1,000 High-Difficulty Clinical Cases (Official gpt-4.1 Grader)

n = 1,000
Asha Apex (June 2026 build)51.4%
Multi-expert ensemble, June 2026; re-graded with the reference rendering
Asha Apex-J (served)48.7%
Frozen 20 Sep 2026; 95% CI 46.9-50.5; +16.3 points over bare GPT-6 Astra
Asha Mid (June 2026 build)41.8%
Three-model merge (898 scored items)
Bare GPT-5.240.3%
Frontier baseline, May 2026 harness (798 scored items)
Bare GPT-6 Astra32.4%
Apex-J's composer model alone, same items
Asha Base (June 2026 build)29.9%
Single retrieval-grounded pass
Bare Gemini 3.1 Pro20.6%
Frontier baseline, May 2026 harness (800 scored items)

Official gpt-4.1 grader with OpenAI's reference rubric rendering. Frozen Apex-J 48.7% (95% CI 46.9-50.5) vs 32.4% for bare GPT-6 Astra on the same items (+16.3 points, p = 3.8e-64). June tiers re-graded the same way; May bare models graded that way when run.

Clinical Crisis & Rare-Disease Safety

psychosis-bench & FPWR PWS 96-Item Red-Team (Pre-Registered 2026-10-03)

n = 96
Asha Apex-J (with Restraint Gate)99.0% safe
1/96 unsafe; 91.0% safe / 9/96 baseline, Holm p = 0.001 vs ChatGPT
Asha Production Cascade85.0% safe
14/96 unsafe; 95.8% safe on psychosis-bench vs 30.2% bare model
Bare Gemini76.0% safe
23/96 unsafe
Bare ChatGPT73.0% safe
26/96 unsafe

Pre-registered rare-disease (Prader-Willi syndrome) and acute psychiatric crisis red-team evaluation. Holm-adjusted p = 0.001 vs ChatGPT.

How the healthcare APIs compare

FeatureAsha APICortiGlass HealthFrontier APIs
Built forMedical Q&A with sourcesDocumentation and codingClinician CDS appGeneral purpose
Free tier250 queries / mo$150 creditsClinician trial$0 to $5 credit
Price$0.03 / query$4 in, $16 out / 1M tokens$20 / mo seat$1.50 to $75 / 1M tokens
Cited medical literature144M+ passages, PMID-checkedNo literature corpusCDS appUnverified, fabricated PMIDs
Public developer APIREST, OpenAI-compatibleRESTApp onlyVendor SDKs
Agent toolingMCP, A2A, CLINo MCP or A2ANoneVendor tools
Crisis safety (PWS red-team)99.0% safeNot testedNot tested24% to 27% unsafe

Asha API

Built for
Medical Q&A with sources
Free tier
250 queries / mo
Price
$0.03 / query
Cited medical literature
144M+ passages, PMID-checked
Public developer API
REST, OpenAI-compatible
Agent tooling
MCP, A2A, CLI
Crisis safety (PWS red-team)
99.0% safe

Corti

Built for
Documentation and coding
Free tier
$150 credits
Price
$4 in, $16 out / 1M tokens
Cited medical literature
No literature corpus
Public developer API
REST
Agent tooling
No MCP or A2A
Crisis safety (PWS red-team)
Not tested

Glass Health

Built for
Clinician CDS app
Free tier
Clinician trial
Price
$20 / mo seat
Cited medical literature
CDS app
Public developer API
App only
Agent tooling
None
Crisis safety (PWS red-team)
Not tested

Frontier APIs

Built for
General purpose
Free tier
$0 to $5 credit
Price
$1.50 to $75 / 1M tokens
Cited medical literature
Unverified, fabricated PMIDs
Public developer API
Vendor SDKs
Agent tooling
Vendor tools
Crisis safety (PWS red-team)
24% to 27% unsafe

Full comparison and methodology: Benchmarks and comparisons.

Pricing

Start free. Pay as you grow.

Start free with 250 queries/month, top up $10 pay-as-you-go to unlock 1,000 evaluation queries/month through January 2027, or pick a monthly plan. Every tier runs the same evidence-grounded Asha engine.

Free Sandbox

$0
  • 250 queries / month
  • 50 queries / day (20 RPM)
  • Up to 2,000 in + 2,000 out tokens / query (8K chars)
  • No spend cap
  • Depth: base

Developer

$49 / month
  • 1,000 queries / month
  • 250 queries / day (60 RPM)
  • 2K in/out included; +$4/1M in, +$16/1M out overage
  • $100 spend cap
  • Depth: base

Pro

$199 / month
  • 10,000 queries / month
  • 2,000 queries / day (120 RPM)
  • 2K in/out included; +$4/1M in, +$16/1M out overage
  • $500 spend cap
  • Depth: base

Enterprise

$999 / month
  • Unlimited
  • Unlimited (240 RPM)
  • Custom volume & BAA
  • $5,000 spend cap
  • Depth: base or auto (Mid $0.30 / Apex $0.85)
Contact sales

Every query includes up to 2,000 input and 2,000 output tokens. Longer documents are metered at $4 per 1M input and $16 per 1M output tokens. Token metering details

Limits, rate windows, and error codes: Pricing, limits, and errors. BAA: read, accept, and check it from the API or asha baa accept, see BAA, self-serve. Higher volumes: [email protected].

Docs

Documentation

Every detail as a table. Every page copies as Markdown, for you or for your agent.