About Self Research IP & Patent Contact

Get in touch

hello@thegeometrical.com
GEOMETRICAL

Category

Episodic Memory System Substrate

Architecture

Hypervisor-level · Local-first · Zero-UI

Every AI you use
operates blindly.
It doesn't know
who you are.

01 · Dissociative Amnesia

No belief system. No "why."

Not your decisions from yesterday. Not why you made them. Not your Gmail history, your Netflix habits, your criteria, your intentions, your roles. Not who you want to become.

02 · Lacks Real Depth

You are never one-dimensional.

You are a founder, a parent, a partner, a learner — simultaneously. Agents treat you as a flat character. None of them can do otherwise. There is no substrate to tell them who you are.

03 · No Primitive

$500M+ raised. Zero local substrates built. The quadrant is empty.

Every agent assumes someone else built the memory layer. Nobody did. That is what Geometrical builds.

The missing primitive

Geometrical
builds the
Episodic Memory
System Substrate.

We operate at the intersection of technology, privacy, and radical personalization.

04

Four surfaces.
One core engine.

Hypervisor-level · System Substrate

S-01

Self

Personal Memory Substrate

The episodic memory of the individual. Captures intention, decisions, criteria — not facts. Operates across all roles. Zero-UI First.

Piloting with beta users

S-02

Soma

Institutional Memory Substrate

The 9-dimensional model applied to collective entities. Organizational knowledge that survives team changes, restructuring, and time.

Piloting with partners

S-03

Axis

Operational Memory Substrate

Memory substrate for autonomous agents. Agents read — log write — the relevant dimensions of a user's identity graph within the Nexus permission boundary.

Piloting M2M

S-04

Nexus

Permission Protocol

Governs what memory is shared between surfaces. Federated graph with user-controlled perimeter definitions. Open source.

In development

AI-Native Architecture

Where Geometrical lives
in the new AI stack.

Without The LLM receives everything or nothing → hallucinates or ignores.
With Geometrical The LLM receives exactly what it needs → reasons correctly.
Agents Claude · GPT · Gemini · Custom Agents Consume Context
↑↓   Minimum Viable Context — served by Geometrical
LLM Inference Ollama · llama.cpp · Core ML · Neural Engine Reasoning
↑↓   Reasoning input — prepared by Geometrical
Geometrical Self · Soma · Nexus · Axis — infers · compiles · preloads context The Substrate Layer
OS Runtime Apple Intelligence · Windows Agent Hub · Android Orchestrate
Hardware Apple M-Series · Snapdragon X · NVIDIA Spark · RISC-V Neural Engine · NPU

Model-agnostic · OS-agnostic · Hardware-agnostic · Zero cloud dependency

AI-Native Device Architecture Ready

Architecture, not philosophy

Why not
cloud.

Cloud memory means every agent interaction transits an external server. At scale, that is a compliance liability in every regulated industry — GDPR, HIPAA, CCPA, EU AI Act. Not a risk to manage. A structural blocker.

Local-first is not a privacy statement. It is the only architecture that passes enterprise procurement without a legal waiver. The data never leaves the device — because it physically cannot.

60%+ of enterprise AI adoption blockers trace to data residency and sovereignty concerns.
Matic Research, 2024

The cloud-first memory players are building into that wall. Geometrical is not.

Why retrieval is not enough

Self vs.
Retrieval Systems.

Retrieval systems find facts similar to the query. Self captures the structure beneath facts: intention, decision criteria, causal chains, role context. These are different problems.

38%

vs 14.4% recall@5
Keep It InMind baseline

Dimension Retrieval systems Self · Geometrical
What it captures Facts similar to query Intention · criteria · causality
Role awareness None Multi-role simultaneous
Temporal structure Flat vector store 4D living substrate
Privacy model Cloud-dependent Local · mathematical
Recall@5 benchmark 14.4%
Keep It InMind
38% direct
+ 12% indirect

Keep It InMind benchmark: Li et al., 2024. Self empirical results: single-user PoC, Aug 2026.

arXivPersonal Episodic Identity Substrate (PEIS) — forthcoming · GitHubAIX Protocol v0.2.0 — Apache 2.0 · PatentP202631047 — OEPM — Priority 19 Jul 2026 · arXivAIX Protocol companion paper — pending · Presshello@thegeometrical.com · arXivPersonal Episodic Identity Substrate (PEIS) — forthcoming · GitHubAIX Protocol v0.2.0 — Apache 2.0 · PatentP202631047 — OEPM — Priority 19 Jul 2026 · arXivAIX Protocol companion paper — pending · Presshello@thegeometrical.com ·

Research & publications

Zenodo · Published

Personal Episodic Identity Substrate (PEIS): A 9-Dimensional Architecture for Human Memory in the Agent Economy

cs.AI · Geometrical Research · 2026 · zenodo.org/records/21923601
Zenodo · Published

The AIX Protocol: A Machine-Native Format for Contextual Identity in Multi-Agent Systems

cs.AI · Geometrical Research · 2026 · zenodo.org/records/21923809
arXiv · Forthcoming

The AIX Protocol: A Machine-Native Format for Contextual Identity in Multi-Agent Systems

cs.AI · Geometrical Research · 2026
Patent · Filed

P202631047 — Privacy-native memory encoding for LLM inference pipelines. Two original inventions.

OEPM · Priority 19 July 2026 · PCT/EPO deadline July 2027
Press · Forthcoming

Coverage, interviews, and media appearances — updates as published.

Press → hello@thegeometrical.com
Research · Forthcoming

IP technical detail paper — privacy architecture deep dive, following PCT filing.

cs.CR + cs.AI · Geometrical Research · 2027

The agents
know everything
about the world.
Nothing about you.

Build
something
that lasts.

We are building with a small number of partners. If you are working on the agent stack, investing in infrastructure, or want to pilot Soma — reach out.

hello@thegeometrical.com