Microsoft Healthcare Agent Service

What is it, and who builds on it?

Microsoft's overview calls it a cloud platform that lets developers inside healthcare organizations build and deploy compliant generative AI healthcare copilots. It has what Microsoft calls a healthcare-adapted orchestrator, powered by large language models. The intended users are IT developers at providers, pharma companies, telemedicine firms and payers. Microsoft's named examples include pharma clinician copilots, hospital protocol question-and-answer bots and administrative workload.

The product has changed names. Microsoft Healthcare Bot became Azure Health Bot in 2021, per InfoQ, and the current documentation titles it the healthcare agent service while URL paths still say health-bot.

What does it do?

Answers can be grounded in the customer's own sources, such as Azure AI Search indexes, websites and documents, and in built-in credible sources. Two layers of safeguards apply. Healthcare Safeguards cover evidence detection, provenance tracking and clinical code validation. Chat Safeguards add a clear disclaimer, evidence attribution, feedback mechanisms and abuse monitoring.

The service also includes a built-in triage and symptom checker, a scenario editor for custom flows, and text or voice self-service. Data connections reach customer systems, such as those that manage EMR data, through generic connectors with no EHR named. On privacy Microsoft says data is protected to HIPAA standards.

What does the "not a medical device" disclaimer mean?

Microsoft's healthcare overview says its services are not intended or made available as a medical device and are not designed or intended to be used in the diagnosis, cure, mitigation, monitoring, treatment or prevention of a disease. The FDA's 510(k) records list no clearance for the service itself, and the two Microsoft entries there are unrelated radiology software products (K173420 and K120734). Clinical and regulatory responsibility for a finished bot therefore sits with the customer that builds it.

How is it priced?

Since early November 2025 the Agent tier bills every Action at $0.01, pay as you go. Actions cost different amounts depending on what the bot does:

• a classic message uses 1 Action
• a customer source or customer extension uses 2
• a credible source uses 5
• a triage session uses 18

At those rates a triage session costs $0.18, a credible-source answer $0.05, and 10,000 triage sessions $1,800.

A Free F0 tier is meant for evaluation and development, with a rate limit of 15 requests per second. The Agent tier allows up to 50. The older S1 tier accepted no new agents after early November 2025. Azure's price page labels figures as estimates and offers a pricing quote, and it links to a free Azure account for trials.

The verdict

This is a build-it-yourself platform for a health system or vendor with developers on Azure, and the per-action pricing makes the cost of a pilot easy to estimate. It is not a ready-made symptom checker for a clinic with no engineering staff, and Microsoft does not sell it as a clinical device.