Research & Development

GraphMemory.
Relational Intelligence.

A local-first behavioral intelligence system. Deterministic pattern detection across human relationships. Cryptographic privacy architecture. No cloud. No account. No server.

Status  In Development · Research Phase
Architecture  Local-First · AES-256
Privacy  Zero Cloud · Zero Account
Pipeline  Deterministic · Psychology-Sourced
Architecture

Three Layers. One Graph.

GraphMemory builds a living map of your relational world from raw interaction logs. The graph is the product. The journal is the input. The revelation is the output.

Cryptographic Identity
256-bit key generated client-side. Never stored. Never transmitted. Every entry encrypted at rest. The key is the only access point — and it belongs entirely to the user.
Pattern Library
A versioned, psychology-sourced behavioral pattern library. Deterministic keyword matching weighted by severity. Cross-referenced against published clinical frameworks. No inference. No guessing.
Relationship Graph
Nodes are people. Edges are shared behavioral patterns. The graph surfaces cross-relationship connections invisible to linear journaling — the same pattern appearing across multiple people simultaneously.
Storage
SQLite · Encrypted at Rest
Local database. Versioned schema with migration path from day one. The user owns the file. Portable, exportable, fully theirs.
Runtime
Tauri · Rust · React
~8MB binary. Native feel. Python analysis sidecar for the pattern pipeline. No Electron. No overhead. Fast enough to open in an activated state.
Export Format
.graphmemory · Context Pod
Encrypted backup format. Portable across installs. Context pod export for AI-assisted analysis — paste into any model. No integration required.
Privacy Model
Zero Knowledge · Zero Telemetry
No account. No server. No analytics. The system has no knowledge of its users. Loss of the cryptographic key means the data is unrecoverable by design.
Research Axis

Active Research Domains.

GraphMemory sits at the intersection of behavioral pattern recognition, longitudinal relationship indexing, and cognitive preservation. The system generates structured, timestamped relational data — a substrate with applications well beyond its consumer interface.

The dementia application is primary. Longitudinal interaction logs as a cognitive continuity record. A structured life archive — encrypted, exportable, transferable to a trusted custodian via key-sharing. The record persists when memory does not.

The semantic layer maps raw interaction language to sourced clinical frameworks — Bancroft, van der Kolk, George Simon, Cialdini, attachment theory literature. Pattern detection with provenance. Every match traceable to a published source.

The Context Pod is the AI interface layer: a structured plain-text export carrying interaction summaries, pattern breakdowns, and health scores — paste into any large language model for reasoned analysis. The pod travels with the user. It requires no integration.

Export Format · Context Pod Header GRAPHMEMORY CONTEXT POD
Generated: [timestamp]
Key fingerprint: [first-8-chars]****
Interactions: [n] · Health score: [0–100]
Patterns: [weighted breakdown]
Red flags: [auto-detected triggers]
Research Domains
Relational pattern recognition
Longitudinal behavioral indexing
Cognitive preservation architecture
Early-stage dementia memory continuity
Semantic experience mapping
Deterministic psychology-sourced pattern detection
Cross-relationship behavioral correlation
Cryptographic privacy in personal intelligence systems
AI-assisted relational reasoning via context export
The graph reveals what linear memory cannot.

When the same behavioral pattern surfaces across three separate relationships — a parent, an ex, a current colleague — the map stops being data. It becomes a life insight. That moment is what GraphMemory is built to produce.

Request early access or research collaboration →

GraphMemory is a service of the Global Data Registry  —  open provenance infrastructure for the machine-readable web.
View the Registry →