HugoScore hugoscore.org

Full review

Savva CAIHL draft report

Evidence-linked HugoScore draft report for a health AI tool that affects patients.

Agency checklist · October 30, 2026

Expands 10 of 12 yes

Document-based draft. Independent confirmation is incomplete. Hugo has not personally verified this service.

Assessed function: Local record app with on-device AI and separately selected cloud-provider queries. Shared API credentials do not make record contents anonymous.

Initial Anthropic coding has an Anthropic model in the role record. Astra review conflicted: OpenAI model in role record.

A no can mean missing documentation or institution-dependent control; it does not by itself establish harmful behavior. These categories are not clinical safety or product-quality ratings.

  • Public documents do not establish actual product behavior.
See all twelve answers and their evidence
QuestionAnswerEvidence / reviewSource and reasoning
1. Told it is AIYesdocumented
Confirmed at document level
the app shows you what will be sent, names the provider, and asks you to agree Source

Privacy policy section 12: cloud AI is off until turned on, and a first-use consent screen names the provider. The home page also labels responses as AI-generated.

2. Operator and model maker namedYesdocumented
Confirmed at document level
The available providers are OpenAI, Anthropic, Google, xAI, Alibaba Cloud, Mistral AI, DeepSeek, and Moonshot AI. Source

Operator is named as Vircode, Inc. (Savva) in the same policy. Cloud model makers are listed by company. On-device models named on the how-it-works page are Llama and Gemma.

3. Patient chooses to use itYesdocumented
Confirmed at document level
Cloud AI is off until you turn it on. Source

The user installs the app and opts in to each cloud provider. No clinician, insurer or employer applies it.

4. Patient sets the taskYesdocumented
Confirmed at document level
12+ AI models on tap. Ask any of them anything. Source

The patient types the question and chooses the model. The Pro tier also uses the chosen AI to structure imported records, which is a fixed operator-side step.

5. Patient can decline or stopYesdocumented
Confirmed at document level
Cloud AI is off until you turn it on. Source

Self-started tool rule: Q3 is yes and no care or service depends on it. The free tier works without AI. Source is the Q3 source.

6. Patient sees the AI outputYesdocumented
Confirmed at document level
AI-powered medical record summaries Source

Listed Pro feature. AI chat answers are shown to the user in the app. Privacy policy says the generated answer is held encrypted for a few minutes so an interrupted reply can be completed.

7. Patient can correct or challengeNonot-documented
Confirmed at document level
Quotation not recorded; evidence repair pending.

No documented route to correct or contest an AI answer or summary. The privacy policy lists a generic right to correct inaccuracies, but Savva holds no health data to correct, and it is not tied to AI output.

8. Patient can get a copyYesdocumented
Confirmed at document level
All charts, all goals, all exports. Source

Free tier includes exports. Privacy policy fn 2 says the only copy is on the device in a portable format.

9. Patient can deleteYesdocumented
Confirmed at document level
You can request that we delete your personal data. Source

Health data lives only on the phone, so the patient deletes it by removing the app. The policy says the cloud AI path keeps no questions, conversations or attachments, and the answer is deleted after a few minutes. Billing records are kept up to 10 years.

10. Training or sale does not override patient choiceYesdocumented
Confirmed at document level
we do not permit the providers to use it to train their models Source

Policy section 12 also says Savva does not use the data for advertising, profiling or sale, and every provider has a DPA with a no-training prohibition. Contractual claim, not independently verified.

11. Patient can run it or take it elsewhereYesdocumented
Confirmed at document level
Run models like Llama and Gemma entirely on-device with zero cloud exposure Source

Local app. Records and on-device AI run on the phone and work offline. The app is proprietary, not open source, and cloud AI needs Savva's relay and API key.

12. No advertising or data businessNodocumented
Corrected after independent review; accepted by editor
product news or offers from time to time Source

Optional promotional SMS offers count as advertising under the written rule, including with consent.

Initial coding: Claude Sonnet helpers, compiled by Claude Opus 5.5 (Anthropic). Selective review and corrections: GPT-6 Astra (OpenAI). Editor: Hugo Campos. Acceptance of corrections is not product verification.

Checklist rules · Independent review and limitations · Later evidence follow-up

Earlier report

Savva: CAIHL reassessment

September 8, 2026. Published AI-assisted draft.

Savva offers local models and separately selected cloud-provider queries, with model and context choices. Shared API keys do not make record contents anonymous. Optional cloud processing and metadata retention qualify broader local-only and anonymity claims.

Scope and agency posture

Local record app with on-device AI and separately selected cloud-provider queries. Shared API credentials do not make record contents anonymous.

Posture: Potentially agency-expanding.

Axis: 78/100, retained provisionally. No numerical recalibration was performed.

Sources and documented findings

  • The current homepage offers local models and provider-specific cloud consent, model switching and context-size control. It says shared API-key traffic is anonymous and also uses broad no-cloud language despite optional cloud processing. https://www.savva.ai/
  • The privacy policy documents instance/device identifiers, diagnostic and security logs, retained billing records, optional health-data transfers and support exceptions. These qualify the homepage's no-identity and no-cloud slogans. https://www.savva.ai/legal/privacy
  • Terms retain a vendor-controlled content framework and refer linked health-data processing to the privacy policy. No hands-on export or output-license resolution was established. https://www.savva.ai/legal/terms

Patient authority

Inference from the documented workflow: Choice of models, local execution and context size give users meaningful control over analysis. Switching models does not itself establish a way to correct extracted records or resolve conflicting advice.

Critical capacity

Editorial assessment: Comparing interpretations can prompt useful questions when users can inspect original records. Agreement between models is not validation.

Informed control

Editorial assessment: An API key identifies the application, not necessarily the patient, but health-record content may identify them. Provider choice must include that distinction and policy retention exceptions.

Assessment and limits

Keep the potentially positive placement while correcting overly absolute anonymity and local-only claims. Current marketing is more developed than the pre-launch baseline, but actual availability was not installed or tested.

Confidence: Moderate for current documented choices, limited for runtime behavior.

Remaining uncertainty: Actual local/cloud boundaries, export, derived-data correction and provider retention remain unverified

Published documentation is evidence of stated conditions, not proof of actual implementation. Unknowns did not receive automatic negative points. Funding, sponsorship, and public code do not determine agency by themselves.

Review provenance

  • Reviewer/model: OpenAI Codex / GPT-6.
  • Method: Focused public-source reassessment using CAIHL: patient authority, critical capacity, and informed control. Existing evidence plus one focused primary-source pass and at most one targeted follow-up. No live product testing. Numerical scores remain provisional editorial placements, not a new calculation.
  • Human review: Hugo Campos authorized publication of these AI-assisted draft reassessments on September 8, 2026. This does not claim comprehensive human verification of every finding.
  • Earlier review: 2026-06-30. Historical assessment. Earlier claims are not automatically reverified by this publication.