Patient-controlled health records AI
Savva
Savva is a beta patient-directed health app that combines medical records, labs, medications, visit notes, wearable and fitness data, and multiple AI perspectives to explain health information in plain language. The vendor privacy policy says health data stays on-device by default, with optional cloud AI inference sent only to a user-selected third-party provider. A CAIHL review should still examine source traceability, correction/export/deletion controls, AI-provider egress, and whether the tool supports agency without drifting into unsupported medical advice. As of June 2026 the app is beta-gated and intended for users 18 and over.
Public-source research has been drafted; final human publication review and change-log detail are still required.
Summary judgment · 78% toward patient-directed
Potentially agency-expanding
A patient-chosen longitudinal record and wearable-data copilot may support reflection and preparation. The local-first design is agency-positive, while optional AI-provider routing, billing logs, correction rights, and clinical boundaries still need review.
Patient agency
How this tool changes agency
Longitudinal records, labs, wearable trends, and plain-language explanations may support reflection and visit preparation if uncertainty and boundaries are clear.
Vendor documentation states each data connection is opt-in, cloud AI requires explicit provider selection, no account is required by default, and on-device processing is central. Hands-on review of onboarding, export, deletion, and subscription flows is still needed.
Patient-facing signals
Who does this AI serve?
Public materials position Savva as a consumer app for people to bring their own records and wearable data into one AI-powered health view.
Can patients tell AI is involved?
Multiple AI perspectives and AI-generated wellness insights are explicit product claims.
Can patients meaningfully choose?
Vendor documentation states each data connection is opt-in, cloud AI requires explicit provider selection, no account is required by default, and on-device processing is central. Hands-on review of onboarding, export, deletion, and subscription flows is still needed.
Can patients correct or challenge what the AI produces?
Needs review of source linking, correction workflows, AI-output feedback, and handling of conflicting model interpretations.
Does it help patients understand or act?
Longitudinal records, labs, wearable trends, and plain-language explanations may support reflection and visit preparation if uncertainty and boundaries are clear.
Text findings
Who is left out or burdened?
Partially documented
The platform is beta-gated, not intended for users under 18, and assumes smartphone access plus Apple Health, Google Fit, EHR, or wearable data connections. Language, disability access, cost, and health-literacy support still need review.
What happens to patient data?
Documented by vendor policy, not independently verified
Privacy policy, last updated February 12, 2026, says connected health data remains on the device by default, no account is required, and Savva does not sell or share data. Optional cloud AI inference sends selected fitness or medical records to the user's chosen provider, including OpenAI, Google Cloud, Alibaba Cloud, Moonshot AI, Anthropic, Mistral AI, or xAI. Savva says its server facilitates transfer but does not save chat history or medical information, while usage is logged for billing. Independent verification is pending.
Are the clinical boundaries clear?
Partially documented
Public site language says AI-generated wellness insights are informational only, not medical advice, diagnosis, or treatment, and may be wrong. In-app boundary behavior, escalation language, and presentation of disagreement across AI models still need hands-on review.
Who defined what good looks like?
Not researched
Needs review of patient involvement, independent evidence, safety testing, and source-grounding accuracy.
Review method
Initial seed classification updated with a 2026-06-12 factual audit of official product pages and privacy policy. No hands-on testing or full CAIHL review.
Draft profile · Low draft, partial public-source check