Cauchy
§00Cauchy / Home — Thesis 2026

Your home runs
its own private AI.

Cauchy ships an open-source small LLM, pre-loaded on a Mac mini that sits quietly in your home. It learns from your calendar, your photos and your sensors — and keeps every audio and video signature inside the walls of the house. Frontier models are invited in through MCP. Your household never goes out.

Request Runtime Access→Read the Manifesto
Reasoning layer ⟶ interchangeable
ChatGPTClaudeGrokGeminiApple IntelligenceTesla OptimusVision ProMeta Ray-BanHumaneChatGPTClaudeGrokGeminiApple IntelligenceTesla OptimusVision ProMeta Ray-BanHumane
§◇The Hub

One quiet box.
The entire house thinks.

Cauchy Runtime begins on a Mac mini sitting silently in a closet. From there, it orchestrates lighting, climate, security, audio, appliances — every connected surface in the home.

Cauchy Hub — a Mac mini at the center of a glowing isometric diagram of household rooms, sensors and devices
Fig. 01 · Cauchy Hub — v0 deployment topologyLocal · Private · Persistent
§01Core Thesis

The future of computing is ambient.

Users will interact with AI through phones, watches, glasses, vehicles, earbuds, robots, and voice interfaces. The AI itself becomes interchangeable.

But physical environments require something the models cannot provide: persistent memory, local context, orchestration, permissions, and environmental state continuity.

Cauchy is the runtime that persists while the interface changes.

§02The Stack

We sit between the model and the matter.

Cauchy preserves continuity while frontier models evolve. The same way operating systems persist across CPU generations.

L3

AI Reasoning Layer

ChatGPT · Claude · Grok · Gemini · future frontier models
L2

Environment Runtime Layer

Cauchy Runtime — memory, orchestration, permissions, multimodal context
L1

Physical Systems

Homes · Vehicles · Buildings · Sensors · Humanoid Robotics
§◈Surfaces

Any interface.
Same runtime beneath.

Cauchy doesn't replace your devices. It gives them somewhere to think — a single persistent context every surface reaches into.

Apple Watch interacting with the Cauchy Hub — A glance. A scene.
Apple Watch

A glance. A scene.

A wrist tap dims the living room, queues recovery music and suppresses notifications — Cauchy reasons the rest.

iPhone · ChatGPT interacting with the Cauchy Hub — Speak in intent.
iPhone · ChatGPT

Speak in intent.

Type or speak naturally inside ChatGPT. The Cauchy Agent routes the action through the household's memory graph.

Laptop · MCP interacting with the Cauchy Hub — Power-user surface.
Laptop · MCP

Power-user surface.

Frontier models call Cauchy's MCP tools directly — your environment becomes a first-class function in any agentic workflow.

AirPods interacting with the Cauchy Hub — Ambient conversation.
AirPods

Ambient conversation.

Walk between rooms and just talk. Path lighting, climate and audio re-balance as you move.

Vehicle interacting with the Cauchy Hub — Arrival, orchestrated.
Vehicle

Arrival, orchestrated.

The car's screen speaks to the home from twelve minutes out — pre-conditioning, garage, lighting, kettle.

Family · Voice interacting with the Cauchy Hub — One household, one brain.
Family · Voice

One household, one brain.

Every voice surface in the home — HomePod, AirPods, phones — funnels into the same persistent context.

§03The Problem

Smart environments today are primitive.

Stateless. Reactive. Ecosystem-locked. None of them can deeply understand a household, hold long-term context, or coordinate the embodied AI systems already arriving.

Today
Alexa Routines
rule-based, cloud-bound
Today
Apple HomeKit
no reasoning, weak context
Today
Google Home
fragmented ecosystem
Today
Home Assistant
powerful but technical
§◉Sovereignty

Your household,
off the cloud sandbox.

Cauchy is sold as an appliance: a Mac mini running an open-source LLM that we pre-load, harden and update. The intelligence lives where the life lives — inside the walls of your home.

The cloud sandbox

Where your home lives today.

  • Apple HomeKit
    Voice + camera streams routed through iCloud relays.
  • Amazon Alexa
    Wake-word audio retained and used for ad and model training.
  • Google Nest
    Faces, voices and presence indexed into a Google identity graph.
  • Ring / Cloud DVR
    Continuous video uploaded; law-enforcement portals attached.
The Cauchy posture

Where your home lives with us.

  • Open-source LLM, on your hardware
    A small, auditable Hugging Face model pre-loaded on the Mac mini we ship. No vendor lock — you can swap or fine-tune it.
  • Audio + video stay in the room
    Microphones and cameras terminate at the Hub. Inference is local. Recordings never touch a third-party cloud.
  • Learns from your own sources
    Calendar, photos, documents, sensors — ingested locally into a memory graph only your household can read.
  • Frontier models on a leash
    ChatGPT, Claude, Grok and Gemini reach in through MCP with scoped, permissioned tool calls. They never see raw signals.
§04Product Architecture

Four components.
One operating context.

See full architecture →
Module · ALocal · Persistent

Cauchy Runtime

The environmental brain. Runs locally on Mac mini, AI PCs, and future Cauchy hardware. Maintains state, memory, multimodal context, automation orchestration and policy enforcement.

  • Environmental state
  • Household memory graph
  • Multimodal processing
  • Local inference
Module · BMCP-compatible

Cauchy Agent

The interface layer. Exposes tools, memory and environmental actions to ChatGPT, Claude, Grok, Gemini and any future frontier model — naturally, from any AI surface.

  • Tool routing
  • Contextual retrieval
  • Action execution
  • Permission scoping
Module · CThe Moat

Memory Graph

Persistent contextual memory of people, rooms, objects, preferences, routines, relationships, environmental patterns and spatial topology. The proprietary substrate that compounds.

  • Spatial topology
  • Behavioral models
  • Object memory
  • Relational context
Module · DEdge AI Appliance

Cauchy Hub

Dedicated edge AI hardware. Initial deployment on Mac mini; future versions with custom AI accelerators and robotics-compatible compute fabric.

  • Local accelerators
  • Sensor ingestion
  • Robotics I/O
  • Privacy by design
§◆The Whole Picture

Agentify your home.
One diagram.

Runtime on a Mac mini. Agent reachable from every frontier model. Memory graph that stays local. Every connected surface in the home, unified into a single intent layer.

Cauchy Home infographic — Mac mini runtime, local memory graph, MCP agent connected to ChatGPT, Claude, Gemini, Grok, HomeKit, Alexa, Google Nest, Philips Hue, wearables, vehicles and robotics
Fig. 03 · Cauchy Home — Agentify diagramUnified · Secure · Local
§05Live Intent

Users do not program automations.
They express intent.

Scene — Help me recover after a stressful day.
inbound · Apple Watch · wearable stress signal

"Help me recover after a stressful day."

Runtime derived
  • ▸ Lighting → warm 2200K
  • ▸ HVAC → 20.5°C
  • ▸ Music → ambient queue
  • ▸ Notifications → suppressed
  • ▸ Bath → preheated
Scene — Prepare the house for movie night.
inbound · iPhone · ChatGPT

"Prepare the house for movie night."

Runtime derived
  • ▸ Dim living-room scene
  • ▸ Blinds → closed
  • ▸ Surround → cinema profile
  • ▸ Kitchen → idle
  • ▸ Door → guest-mode
Scene — I'm 12 minutes out — start arrival.
inbound · Vehicle · Grok in-car

"I'm 12 minutes out — start arrival."

Runtime derived
  • ▸ Garage → open on geofence
  • ▸ HVAC → pre-condition
  • ▸ Entry path lighting
  • ▸ Kitchen → kettle on
  • ▸ Dog camera → live
§06Future Surfaces

Built for the
physical AI era.

The same runtime, the same memory, the same permissions — extending from today's home into tomorrow's embodied systems.

AR Glasses

Spatial recall

“Where did I leave my passport?” Cauchy queries environmental memory and surfaces a last-known-location overlay through the lens.

Vehicles

Arrival orchestration

Your car speaks to Cauchy. The house pre-conditions, garage opens, evening routine begins — before you turn into the driveway.

Wearables

Biometric continuity

Sleep debt and elevated HRV become inputs. Lighting warmth, caffeine timing and notification intensity adapt across the day.

Humanoid Robotics

Embodied tasks

Cauchy instructs Optimus through orchestration APIs — set the table, organize seating, assist with serving — using the household's spatial graph.

Elder Care

Dignified guardianship

Abnormal movement, missed routines and irregular sleep trigger graduated responses — all reasoning held locally.

Hospitality

Portable preferences

Lighting, temperature, music and room layout persist across hotel stays — the room learns you, not the chain.

§07The Moat

Not prompts.
Not automations.
Persistence.

Cauchy Memory Graph — luminous blueprint of rooms, people and routines connected by glowing filaments
Fig. 02 · Memory Graph topologyUser-owned · Local
→
Spatial memory graphs
What is where, and how it relates.
→
Household behavioral models
How this specific home actually lives.
→
Environmental orchestration
Hundreds of devices, one coherent intent.
→
Contextual continuity
State that survives model migrations.
→
Device topology intelligence
A live map of the physical substrate.
→
Multimodal environment understanding
Vision, audio, sensors, presence.
Private Beta · 2026

The environment persists.
The interface changes.

We are onboarding a small cohort of AI-native homes, vehicles and hospitality partners for the first Cauchy Runtime deployments.

Request Runtime Access →Read the Manifesto