Almanac Health

The AI is the product. Almanac Chat answers clinical questions from vetted clinical sources with citations, works with patient context from a connected record, analyzes uploaded research and data, runs code, and drafts summaries, presentations and patient education materials; the health system offering is the same platform 'designed to work within existing electronic health record (EHR) systems'. Remove the model and no reference database, workflow tool or record is left to sell.

How a clinician oversees the assistant is described; where that oversight stops is not. Every answer comes with citations 'so you can check the evidence yourself', and the terms require the clinician to review all outputs for accuracy and completeness and never to use output about a patient for clinical, treatment or diagnostic decisions 'without appropriate verification and professional assessment'.

The product also acts on the clinician's behalf: it connects to the record, runs code on uploaded data, and drafts documents and patient education materials. Nothing published says what the assistant may not do, whether any action it takes in a connected record needs the clinician's approval, or what happens when an answer or a computation is wrong. The line that every answer is grounded and cited describes a mechanism a clinician can check; it is not a claim that answers are right.

The approach is named without the model or version behind it. Almanac describes answers grounded in vetted clinical sources and peer-reviewed literature, with citations, and its April 2026 announcement ties the product to the founder's Stanford work on retrieval-augmented generation for clinical medicine published in NEJM AI. The terms mention 'our models, algorithms, or systems' and 'Third Party Services' whose output may appear in the product. No foundation model or provider is named, the source corpus is not characterized beyond 'vetted clinical sources', and no versioning or change practice is published.

Almanac describes its architecture but names no model provider. The privacy policy (last updated 26 January 2026) says Almanac shares information with 'providers of hosting services, cloud services, and other information technology providers', naming none, and processes it in the United States; the terms say parts of the service may include output from Third Party Services under their own terms, which leaves open whether a third party model receives clinical questions and patient context. No subprocessor list is on the open site, though the Vanta trust center at trust.almanac.chat may carry one.

The headline claim is 'research-validated' and 'clinically validated through peer-reviewed research', and what the site points to is the founder's earlier Stanford work on retrieval augmented generation for clinical medicine, published in NEJM AI, a study of a research system rather than of Almanac Chat as sold.

The April 2026 announcement says the platform 'is currently undergoing rigorous clinical validation in academic medical center settings', naming no institution, and no customer or deployment is named in anything Almanac publishes. The App Store listing carries five ratings. The product is early, with a strong research lineage, and no validation protocol, sites or first results are published.

A training position is published, and it includes a grant over some users' content. The terms say 'By default, we do not use the Content of registered users to train our systems', with an opt-in through feedback or account settings, while public users' content may be used for training unless they opt out; content is also used to 'provide, maintain, develop, and improve our Services'.

The privacy policy (26 January 2026) says patient data in business offerings is governed by business associate agreements, and retention of personal data is 'as long as we need', with no period stated. Nothing states how long questions, uploaded files and record data are kept or whether anything is deidentified. Safety engineering, meaning guardrails on generated clinical answers, hallucination mitigation and how a safety event is handled, is not published beyond citations.

Business associate status is stated and supported, and its scope across plans is not. The privacy policy (last updated 26 January 2026) says content processed for customers of Almanac's business offerings falls outside the policy and that 'patient data is governed by Business Associate Agreements'; the App Store listing calls the product 'a HIPAA-compliant AI workspace' and invites clinicians to connect their record; the terms forbid users to violate HIPAA. What is not published is the agreement, and whether an individual clinician on the Free, Pro or Max plan who connects a record is covered by one, which is the entry buyer's question.

No independent assurance is shown on the public site. The home page links a trust center (trust.almanac.chat, hosted on Vanta) that could contain a report or certification. Elsewhere the pages assert that institutional data remains secure and that the workspace is HIPAA compliant, and name no SOC 2 report, ISO/IEC 27001 or HITRUST certification or penetration test.

No clearance is claimed and no regulatory position is stated. The company describes itself as 'evidence-based clinical decision support', and the product answers questions on differential diagnoses, treatments and medications, and its site lists treatment verification that 'checks drug interactions, resistance patterns, and institutional protocols'.

The terms say Almanac does not provide medical advice, diagnoses or treatment and that clinicians must verify output before any decision about a patient, and the plans page says it is built for licensed clinicians and trainees, not patients. Each answer shows its sources, so the clinician can review the basis. That is a responsibility clause beside clinician-facing recommendations, not a regulatory analysis.

Governance is committed to in policy language with no evaluation published. Almanac describes its platform as 'governed by institutional controls' and 'free from pharmaceutical advertising', and its April 2026 announcement commits to 'rigorous evaluation before deployment'; the home page says its work is 'disciplined in what we ship and how'. No governance process is described, no evaluation of answer accuracy by specialty or question type is published, and no results by patient group are given. The trust center may hold more.

Source traceability is the mechanism, and nothing stands behind the output. Every answer carries citations a clinician can open, and the latest app release adds inline citations that jump to the source. The terms state that 'Output may not always be accurate', that Almanac 'does not guarantee the accuracy, truthfulness, adequacy, or completeness of any information', provide the service 'as is', cap direct damages at $50 and run indemnity from the user to Almanac. No error or accuracy rate for answers is published, and no route is described for a patient affected by a wrong answer.

Integration is claimed and no system is named. The App Store listing says clinicians can 'Connect your EHR to bring relevant patient information into the conversation' and review medications, labs, allergies, notes and history; the funding announcement says the platform is 'designed to work within existing electronic health record (EHR) systems'. The site names no record system, standard, marketplace listing or production site.

A single hosted service with location implied rather than committed. Almanac runs as a web application with iOS and Mac apps, and the privacy policy (26 January 2026) says users consent to processing of their personal information in the United States and that hosting and cloud providers are used, naming none. Nothing states where patient data from a connected record is processed or stored, how institutions are separated, or whether the individual plans and the health system offering share infrastructure.

Tiers, figures, a unit and a self-serve start are published for individual clinicians. The Almanac Chat plans page lists Free at $0 a month (limited daily messages), Pro at $30 a month and Max from $150 a month for 5 or 20 times Pro's usage, each started from the page ('Try Almanac', 'Get Pro', 'Get Max'); the App Store lists in-app prices of $39.99 (Pro), $389.99 (Pro yearly), $199.99 (Max) and $399.99 (Extended Max). The health system offering is by early access and unpriced. The company also states the platform is 'free from pharmaceutical advertising', which tells a buyer how it is not funded.

Coverage is stated clearly by user and broadly by field, with nothing validating it yet. Almanac is built for licensed healthcare professionals in active practice and for medical students, residents and fellows in accredited programs, not for patients, and it offers decision support 'across medical specialties'. No care setting is characterized, no specialty depth is described, and clinical validation is described as under way in academic medical center settings without results.