Build with us
We scope one workflow with your team, build the software, and support it after launch.
- Workflow discovery with the people who do the work
- Working prototypes your staff can test
- Deployment, monitoring, and support
Healthcare AI, built with you.
We train and develop our own medical language models, then build AI software on them for hospitals, physician groups, and research institutes. Your developers can call the same models through a HIPAA compliant* API.
Three ways to work with us
Start with the one that fits your team.
Council models
Language models we fine tune for medical text. Call them through our API or run them inside your own cloud account.
- Tuned for clinical and biomedical language
- Available through the API
- Deployable in AWS, GCP, or Azure
Developer platform
An OpenAI compatible API with retrieval over your own text and cited answers. HIPAA compliant*, with a BAA available on request.
- Chat completions and streaming
- Knowledge bases with citations
- Scoped keys and usage tracking
Software shaped around your workflow
Choose your organization to see the kinds of tools we build. Every project starts with one workflow and a goal your team can measure.
- Notes and orders
- Council
- Draft summary
- Clinician signs
Discharge summary drafts
Drafts assembled from notes, orders, and results, ready for the attending physician to edit and sign.
Prior authorization packets
Gathers the documentation a payer requests and flags anything missing before submission.
Policy and protocol answers
Staff ask questions of approved policies and get an answer with the source page attached.
Examples of tools we can build. Each one is scoped with your team, and its outputs go to qualified staff for review before use.
How a project runs
Your clinical, operational, and security leads stay involved from the first conversation to launch.
- 1
Scope
We meet the people who do the work and choose one workflow with a clear goal.
- 2
Prototype
Your staff try a working version on sample cases and tell us what needs to change.
- 3
Validate
Clinicians review outputs against criteria you set before anything goes live.
- 4
Launch
We deploy in our cloud under a BAA or inside your cloud account, with access controls and audit logs.
- 5
Support
We watch quality after launch, ship updates, and extend the tool when the work changes.
Models tuned for medicine
Council 1 is our family of language models, fine tuned for clinical and biomedical text. Use them through the API, or deploy them inside your AWS, GCP, or Azure account so patient data stays in your environment.
Read the Council 1 Pro benchmark write upResearch benchmark results on synthetic cases. They do not measure clinical performance.
- Your applicationThe tool we build with you, or your own product.
- Retrieval and citationsAnswers grounded in approved sources, with references.
- Council modelsLanguage models fine tuned for medical text.
- Your cloud or oursYour own account, or our hosted API under a BAA.
Medical models and cited retrieval, through one API
The Council API is HIPAA compliant* and works with OpenAI SDKs. Add your own text to a knowledge base and get answers with structured citations.
from openai import OpenAI
client = OpenAI(
base_url="https://developer.council.health/v1",
api_key="YOUR_API_KEY",
)
response = client.chat.completions.create(
model="YOUR_MODEL_ID", # GET /v1/models lists yours
messages=[{
"role": "user",
"content": "Summarize first line treatment for hypertension.",
}],
extra_body={
"citations": {
"enabled": True,
"sources": ["public_kb"],
"top_k": 3,
}
},
)
print(response.choices[0].message.content)Summarize first line treatment for hypertension.
Guidelines recommend lifestyle changes for everyone with high blood pressure. When medication is needed, first line options include thiazide diuretics, ACE inhibitors, ARBs, and calcium channel blockers [1][2]. The choice depends on age, other conditions, and blood pressure level [1].
- 1Clinical practice guidelinePublic knowledge base
- 2Evidence reviewPublic knowledge base
Sample response for illustration. Not medical advice.
What happens when you ask for citations
Built for protected health information
Patient data calls for conservative defaults. These are the commitments we make before any protected health information is involved.
HIPAA compliance covers protected health information handled under a signed Business Associate Agreement, and it also depends on how each party configures and operates its systems. We walk your security team through our safeguards and your responsibilities before launch.
Read our security practicesCommon questions
AI software for specific clinical, operational, and research workflows, such as documentation drafts, prior authorization packets, chart abstraction, and policy search. We also train medical language models and run a developer API.
Yes. The Council API and the software we deploy are HIPAA compliant* for protected health information handled under a signed Business Associate Agreement. We sign the BAA with you before any protected health information reaches our systems.
Yes. Council models and the software we build can be deployed in your AWS, GCP, or Azure account. An air gapped option is available for environments that need stricter isolation.
No. Your data is not used to train shared models, and by default the API does not keep raw prompts or completions in its logs.
It follows the OpenAI chat completions format, so OpenAI SDKs work with a Council base URL and API key. Add a citations object to a request to ground the answer in a knowledge base and receive the sources it used.
No. They support clinical, administrative, and research work, and their outputs are meant for review by qualified professionals. They are not intended to diagnose or treat patients or to replace clinical judgment.
Tell us about one workflow. We set up a call with the people who own it, then propose a scope, a timeline, and a way to measure success.
Find the right place to start
Answer a question or two, and we will point you to the right next step: an API key, a call with our team, or a quick note.