Marisa.Care

The agents are the product. Marisa.Care sells three AI agents to health institutions: a call center agent that answers patients by voice and WhatsApp in natural language, 24/7, and books, cancels and reschedules against the real schedule with no menu; a concierge agent that finds prescribed exams and consultations never booked, confirms appointments, rescues no shows and runs digital check in; and a clinical support agent that transcribes consultations and drafts structured notes, certificates and referrals. The conversations, the transcription and the drafting are done by the models, and the integration layer exists to feed them. Remove the models and there is nothing left to sell.

The agent is always on and the handoff is a phrase. The call center agent answers patients 24/7 by voice and WhatsApp and books, cancels and reschedules without human intervention; the home page says it transfers to the human team with the conversation's context when the case requires, directs sensitive cases to the human team automatically, and prioritizes patients by mapped health risk. The concierge agent contacts patients unprompted about pending exams from the record.

No criterion is published for what counts as sensitive, what a patient describing an urgent symptom hears, the hours of human cover, or how risk prioritization is set; the Data Protection Notice leaves validation and supervision to each institution.

The architecture is described in general terms with nothing identified. The home page says the agents are orchestrated by the company's own infrastructure, which centralizes and interoperates data and integrations, and that the call center agent works in natural language by voice and text. No speech recognition or language model, version or update practice is named, and no evaluation method is published.

The architecture is described and no model provider is named. The home page describes the company's own orchestration infrastructure, and WhatsApp is named as a patient channel, which places a messaging platform operator in the chain. The Data Protection Notice says data may be shared with service providers and suppliers and processed abroad through cloud services, without naming any of them. No subprocessor list, host, speech recognition provider or language model provider is named.

Vendor reported ranges with no method. The home page reports, as ranges "based on clients with a similar profile", more than 60 percent of contacts resolved by the AI without a human handoff, a 30 to 50 percent cut in operating cost, 15 to 30 percentage points more prescribed exams converted, and a 20 to 40 percent reduction in no shows, and says each operation is projected in a commercial assessment. No customer, period, baseline or denominator is given for any figure, and no study of the agents is published.

General assurances that do not answer the AI questions. The Data Protection Notice says patient data is processed automatically, including with artificial intelligence, to carry out the service, support continuity of care, organize information, support process management and improve the quality of the service, under the partner institution's guidelines; it keeps data as long as needed for the purpose, and offers anonymization, blocking or deletion on request.

It does not say whether conversations, recordings or record content are used to train or improve the agents' models, what is retained after a call, or whether anything is de identified before it reaches a model. No safety engineering is published: no guardrails for a patient facing agent and no account of how a safety event is handled.

A non US regime documented to depth. Marisa.Care is a Brazilian company, and its Data Protection Notice (last updated 10 June 2026) states that when its solutions are used in care delivered by a partner institution, the hospital, clinic, health plan or laboratory is the controller of the patient's data and Marisa.Care acts as operator, processing on the institution's instructions within its contract.

The notice is written under the LGPD, names the data protection officer's channel (encarregado@marisa.care), sets out data subject rights and commits to report security incidents to data subjects and the ANPD. The company does not invoke the United States rule, and business associate terms for United States customers are left unaddressed.

Controls are asserted with nothing independent behind them. The home page describes a robust infrastructure aligned with regulatory standards, and the Data Protection Notice commits to technical and organizational measures and to notifying data subjects and the ANPD of relevant security incidents. No ISO/IEC 27001, SOC 2 or other independent attestation, penetration testing statement or trust page is published.

A scope statement sits beside clinical support. The site footer and the Data Protection Notice say Marisa.Care does not provide medical services or clinical care and that its solutions are technological and operational, supporting institutions' process management, and name a physician as technical lead. The clinical support agent is marketed as offering evidence based decision support during the consultation, and the call center agent prioritizes patients by mapped health risk.

The same agent directs "sensitive cases" to the human team automatically, and nothing published says what counts as sensitive or who decides. No registration or position under ANVISA, the Brazilian device regulator, or any other regime is published. A responsibility statement beside clinician facing support is tension rather than a stated regulatory position.

Governance is committed to in policy language with no evaluation behind it. The Data Protection Notice describes a data protection governance program and says the solutions follow validation and supervision flows set by each institution, and it grants a right to request review of decisions made solely by automated processing, with transparency limited by trade secrets.

No evaluation of the agents is published, including how well the voice agent understands callers across accents, age groups and speech patterns, or how the risk prioritization performs for different patient groups.

Mechanisms exist, and the recourse route runs through the institution. The call center agent hands conversations to the human team with context, and the Data Protection Notice says the solutions follow validation and supervision flows each institution defines.

The notice grants a right to request review of decisions made solely by automated processing, with transparency limited by trade secrets; because Marisa.Care acts as operator for patient data, with the institution as controller, that right runs to the institution rather than being answered by Marisa.Care. No error rate, warranty or remediation commitment for a wrong booking, a missed urgent call or a wrong drafted note is published.

Integration is claimed with no system named. The home page describes native integration with the main hospital information systems on the market and says the agents operate connected to the medical record, the schedule and the institution's systems, reading prescribed exams from the record and booking against the live schedule. No record system, hospital information system, standard or deployment is named.

A single hosted service with its location left open. The agents run as Marisa.Care's hosted service over voice and WhatsApp. The Data Protection Notice says data may be transferred to, stored in, processed in or accessed from other countries through cloud computing, support, communication and security services, under LGPD safeguards such as standard contractual clauses; no provider, region or country is named, and tenancy is not described. WhatsApp as the patient channel adds a messaging platform operator to the processing chain.

No price is published, and the posture is discoverable. Marisa.Care sells to hospitals, clinics, health plans and laboratories, and the route in is a meeting booked through the site, where a commercial assessment projects results for the buyer's operation. The home page frames value as cost reduction and revenue recovered from converted pending exams and filled no show slots. No price, unit of charge or contract shape is published.

Coverage is stated clearly with nothing validating it yet. The agents are sold to hospitals, clinics, health plans and laboratories in Latin America for call center, patient access, follow up and consultation documentation, with WhatsApp and voice as the channels. The outcome ranges on the home page are vendor projections without a named customer or method, so no part of that coverage carries published validation.