# Rhapsody Launches Guardian to Control AI Agent Access to Healthcare Systems

Rhapsody Guardian is now available, bringing scoped access, rate limits and request records to AI agents, MCP connections and healthcare APIs.

By Paras Madan | 2026-10-21


On October 20, 2026, Rhapsody [announced that Guardian is available](https://rhapsody.health/blog/rhapsody-launches-guardian-to-govern-ai-agent-and-api-access-to-healthcare-data-and-services/), introducing an access-control product for AI agents and applications reaching healthcare data and services. Unveiled at its Connect conference in Boston, Guardian authenticates callers, restricts permitted operations, applies rate limits and records the activity passing through it.

The launch addresses a concrete engineering problem. An organization can approve an AI project while still lacking a consistent way to control what the deployed agent touches. A policy document cannot stop an API request on its own. Guardian is designed to put those decisions into the request path, where access can be allowed or blocked before the underlying service responds.

Rhapsody says the product can run independently or alongside Rhapsody, Corepoint or another integration platform. That gives the announcement relevance beyond existing customers of one integration engine, although a buyer still needs to establish how it fits a particular environment.

## Enforcing the boundary around a service

Rhapsody's [governance overview](https://rhapsody.health/solutions/ai-governance/) describes three parts of the workflow. Administrators define callers and the services and operations each can use. Guardian verifies incoming requests against those policies. Teams then review traffic and enforcement records and adjust the rules as requirements change.

A narrowly scoped deployment could, for example, let an approved agent read from a designated service without permitting updates. A request limit could cap how often that agent calls it. That example illustrates the operational value: the access boundary is expressed in the system receiving the request rather than left entirely to the agent's instructions.

This is useful even when the agent normally follows its brief. A planning mistake, retry loop or poorly scoped tool can still produce unwanted requests. The quality of the access policy therefore matters as much as the quality of the model. A gateway can enforce the permissions it receives; it cannot decide the organization's appropriate permissions for it.

## Where Guardian fits in the stack

The [Guardian product page](https://rhapsody.health/solutions/ai-governance/guardian/) lists support for OAuth 2.0, SMART on FHIR and Basic Auth, together with allow/block rules and access logging. Its architecture covers FHIR APIs, REST APIs and MCP requests. The monitoring capabilities include a gateway view, transaction and deployment logs, and graphical metrics.

Guardian also has a defined boundary with the integration layer. It can govern an existing API directly. When a workflow needs data transformation, routing or connections across systems, the integration engine still performs that work while Guardian controls access to the exposed service. Buying the governance layer does not remove the underlying integration work.

That boundary is important for evaluation. A team should test an allowed request, a forbidden operation and a burst above the configured limit, then check whether the resulting records explain what happened. It should also map any routes that bypass the gateway. Visibility into traffic passing through Guardian should not be mistaken for visibility into every action an agent can take elsewhere.

The public launch material does not provide pricing, performance benchmarks or an independent assessment of deployment outcomes. Its claims concern access enforcement and oversight, not proof that an agent's clinical reasoning is correct or that a deployment automatically meets every compliance obligation.

For enterprise AI teams, the significant change is a product built around controlling agent access to existing services. The adoption question is whether those controls match the organization's real systems and whether the records are useful when something needs investigating.

I'm Paras, and I help technical founders make complex infrastructure products easier to understand.

If you're launching an AI governance or enterprise platform, I can help explain the mechanism and customer value without stretching the security claims.

[Book a 30-minute call](https://cal.com/paras-madan/30min).