Sigmastead

Personal data platform · in private development

Your life isn't
one app at a time.

Sigmastead is building a personal data platform that helps people bring selected information together, discover patterns across sources, set goals, and control what they choose to share.

Get in touch Private development · Chester, Virginia

Every app knows a sliver.
Nothing knows the shape.

Your health app knows how you slept. Your calendar knows what you did that week. Your bank knows what it cost. None of them can tell you how those things move together, because none of them can see past their own edges — and none of them are built to.

The question worth answering isn't inside any single service. It's in the relationships between them.

What we're building

Bring it together

Selected data from different parts of your life, in one place you choose — rather than scattered across the services that happen to hold it.

See across sources

Relationships that individual applications cannot reveal, because they can only see their own slice.

Say what matters

A Vision Board and goals, so the system knows what you are trying to improve rather than guessing at it.

Observations with sources

An AI assistant that produces observations you can trace back to the specific evidence behind them, not unattributed claims.

You choose disclosure

Control over which parts of your data or evidence you share, and with whom.

Minimized by design

The analysis path is designed so that raw records, calendar titles, attendees, locations and device names are stripped before anything is sent for analysis.

The intended shape

This is the architecture as designed. It is not a description of a deployed system — see where things actually stand below.

Product architecture in development: sources, ingestion, a normalized boundary, evidence minimization, AI analysis, and user-controlled disclosure Sources the user selects are ingested locally or with authorization into a private normalized data boundary. Inside that boundary, a minimization step selects evidence. Only minimized evidence — a question, opaque identifiers, dates, metric names and aggregate values — leaves the boundary for AI analysis. A source-supported result returns to the user, who decides what to disclose. The diagram describes a design, not a deployed system. Sources you select Health Calendar others, TBD local or authorized ingestion Private normalized boundary Normalized your records Minimize select evidence Stripped before anything leaves: titles · attendees · locations device names · free-form notes only minimized evidence leaves AI analysis designed, not yet run Result with its sources You choose what to disclose
Product architecture in development. The minimization step is the part worth looking at: the analysis payload is designed to carry only a question, opaque identifiers, dates, metric names, aggregate values and units — not your records.

Where things actually stand

Most pre-launch sites describe the finished thing in the present tense. Here is the real state instead, because a company arguing for data honesty should probably start with its own.

Ingestion
A tested local prototype imports one Apple Health export and calendar files, and strips identifying fields at parse time. It is a one-time manual import, not a sync service. No production connectors exist.
Analysis
Designed against Amazon Bedrock with a structured response contract. No analysis request has successfully run. The usefulness of the assistant is unvalidated.
Storage and keys
Undecided. We are not claiming a private vault per user or user-held encryption keys, because neither has been designed to a standard we would defend.
Export and deletion
Designed, not built. We will not describe portability or erasure as available until they work.
Compliance
Nothing is audited or certified. No SOC 2, no ISO 27001, no attestation of any kind.
Users and revenue
None, and none.

Early, and talking to people.

If you work on personal data, cross-domain analysis, or privacy engineering — or you have a concrete problem here and would try something rough — that's the conversation we want.

Get in touch