Private health intelligence layer
Your health data becomes a living map — one that reasons across itself.
CareVault pulls the scattered pieces of your health into one controlled record — labs, biomarkers, medications, wearables, imaging, notes, and care dates — then does what a folder never could: correlates two signals, measures a medication's before/after impact, aligns every series onto one timeline, and reconciles units across sources. It runs on your own machine and works with any AI. No portal archaeology. No silent gaps. No silos.
72 tools · runs on your machine
Cross-signal reasoning · new
Any record can store a lab result. CareVault's newest layer reasons about relationships — the part where health insight actually hides. It aligns any two signals onto a common footing and returns real statistics, always descriptive, never a diagnosis.
correlate_metricsPearson & Spearman between any two signals — weight against A1C, sleep against resting heart rate — aligned on a shared day, week, or month grid, with paired sample size, a two-sided p-value, and honest caveats about small samples and autocorrelation.
analyze_event_impactAnchor a metric to a discrete event — a medication start, a procedure — and get before/after descriptive stats plus a Welch t-test on the change, with an optional washout gap to skip the transition.
align_seriesResample many signals onto one shared time grid — one row per bucket, one column per signal — so any two line up without hand-matching a single timestamp.
normalize_seriesReconcile units and reference ranges across labs and devices — mg/dL ↔ mmol/L via analyte molar mass — and add a unitless in-range position so results from different labs finally compare.
Health trajectory
Every dated detail becomes part of the same path. A medication change can sit beside a lab shift. A sleep baseline can explain a symptom window. A follow-up stops being a memory burden and becomes part of the map.
Record model
CareVault is not a prettier PDF drawer. It gives each kind of health data a structured place to live, then connects the pieces by date, source, and meaning — so the reasoning layer has something real to reason over.
A report is stored as metadata plus individual results, so LDL, A1C, CRP, vitamin D, and other analytes can become timelines.
Units, flags, specimen, lab name, ordering provider, and source notes remain part of the record — ready to normalize across labs.
The MCP can search, summarize, correlate, measure event impact, align, normalize, and export the data without hiding the structure.
Whole-record coverage
The underlying HealthLedger MCP is a local-first, model-agnostic server exposing 72 structured tools: broad capture across clinical history, high-volume wearable samples, care tasks, documents, reproductive records, substance logs, and oncology — plus a cross-signal layer that reasons over all of it.
Lab reports, individual results, flags, units, reference ranges, and descriptive trends.
Active meds, dose logs, schedules, instructions, refill dates, and adherence history.
HRV, sleep, resting heart rate, steps, SpO2, workouts, temperature, CGM-style samples, and scales.
Tumor records, body site, stage, grade, marker summaries, treatment status, and notes.
Encounters, annual visits, procedures, imaging, immunizations, follow-ups, and care tasks.
Reports, notes, attachments, source metadata, search, and exportable record pages.
Family history, reproductive records, substance logs, profile facts, and user goals.
Correlate two signals, measure event impact, align many series, and normalize units and ranges.
Control layer
CareVault treats the assistant as a reader and operator over your record, not the owner of it. The data model stays explicit, exportable, and inspectable — and every analysis stays descriptive.
Labs, doctors, pharmacies, devices, notes, imaging, and documents converge into one private longitudinal record.
search_records → all domainsAny two series align on a shared grid, so a correlation, a before/after change, or a unit reconciliation is one call away.
correlate_metrics → weight × A1CRefills, follow-ups, labs due, stale data, immunizations, and unresolved tasks become a single care agenda.
health_agenda → next 30 daysLocal-first · model-agnostic · open source
Not another portal. Not a hosted account. CareVault is the surface for a personal health ledger you run yourself — a local MCP server that keeps your data in a file on your own machine and works with any AI client and any model. Install it, point your assistant at it, and start reasoning over your own record.
A single-tenant live demo runs at health-mcp.manticthink.com/mcp — to actually use it, self-host your own from GitHub.