Local-First Physical AI Runtime

The execution layer for Physical AI.

Large language models can reason about the world — but they can't act in it. EdgeMind Core is the missing runtime that lets AI agents perceive, reason about, and directly control real-world devices: sensors, robots, Arduino, FPGAs and industrial hardware — fully offline, low-latency, and without rewriting your existing systems.

STATUS: Pilot — validated across multiple hardware scenarios
ORIGIN: National Tsing Hua University · Taiwan
FOCUS: Persistent Physical AI Infrastructure
The Missing Layer

AI can think. It still can't touch.

Today's models have powerful reasoning. But the moment an AI needs to control a sensor, a robotic arm, an Arduino board or industrial equipment, it falls back on engineers writing custom integrations, device protocols and control logic by hand — turning every deployment into expensive, one-off engineering.

Intelligence

LLMs & AI Agents

Capable of language, reasoning, planning and decision-making.

no shared interface weeks → months to bridge
Physical World

Devices & Hardware

Fragmented protocols, custom firmware, isolated control logic.

PROBLEM / 01

Every deployment is bespoke

Custom firmware, protocol handling and control logic make each automation a costly, non-repeatable engineering project.

PROBLEM / 02

SMEs are locked out

Small and mid-size factories face labor shortages and knowledge gaps, but lack the engineering and AI teams to deploy automation.

PROBLEM / 03

Integration time kills momentum

Even simple device coordination takes weeks or months to integrate — keeping AI out of real-world environments entirely.

The real bottleneck for Physical AI isn't model capability —
it's the absence of a standardized execution interface that lets AI generalize across real-world devices.

EdgeMind Core

One runtime. From reasoning to actuation.

A Local-First Physical AI Runtime that gives AI agents a unified capability layer and a modular skill system — so they can sense, reason, and control real devices directly, without rewriting your hardware.

AI AgentLAYER 03 · INTELLIGENCE

Natural-language intent, reasoning and planning from any LLM or agent.

↓ intent
EdgeMind Core
Capability + Skill LayerLAYER 02 · EXECUTION

A unified capability model lets the agent discover what each device can do; modular skills translate intent into safe, permissioned actions — locally, with low latency.

capability_map skill.execute() permission_gate edge_inference
↓ actions / ↑ telemetry
Physical DevicesLAYER 01 · WORLD

Sensors · Arduino · FPGAs · robotic arms · cameras · edge devices.

An execution interface AI can generalize across

Instead of bespoke firmware development or custom system integration, EdgeMind exposes every connected device through a unified capability layer. An AI agent simply asks what a device can do — and a modular skill system handles how it gets done.

The result: AI understands device capabilities and executes real-world tasks from natural-language instructions — monitoring environmental conditions, coordinating robotic actions, or driving industrial equipment — without anyone building custom AI infrastructure from scratch.

Because it's local-first, EdgeMind keeps running where the cloud can't: on the factory floor, in classrooms, in remote sites and low-network environments — with permission-based safety controls on every action.

Core Capabilities

Built for the edge of the real world.

EdgeMind transforms fragmented hardware into AI-accessible systems — designed from the ground up for reliability outside the data center.

/ 01

Low-latency local execution

Inference and control run on-device, so actions fire in real time — no round-trip to the cloud, no jitter on the factory floor.

/ 02

Fully offline operation

A local-first runtime that stays online when the network doesn't — keeping Physical AI persistent in remote and disconnected environments.

/ 03

Modular skill system

Device behaviors are packaged as composable skills, so new hardware plugs into the same interface instead of demanding a fresh integration each time.

/ 04

Permission-based safety

Every action passes through explicit permission gates — giving operators control over what an AI agent is allowed to do in the physical world.

Hardware EdgeMind speaks natively

ArduinoFPGA SystemsRobotic ArmsEdge SensorsCamerasEmbedded ControllersIndustrial EquipmentEdge Devices ArduinoFPGA SystemsRobotic ArmsEdge SensorsCamerasEmbedded ControllersIndustrial EquipmentEdge Devices
Where It Runs

From the shop floor to the field.

The same runtime, the same interface — deployed across environments where conventional AI integration is too slow, too costly, or simply impossible.

USE CASE / 01

Small & mid-size factories

Coordinate machines, monitor conditions and automate workflows without an in-house AI team — closing the labor and knowledge gap that holds SMEs back.

USE CASE / 02

Education & makers

Students and makers control real hardware through natural language — turning Arduino, sensors and robots into a hands-on Physical AI playground.

USE CASE / 03

Remote & low-network sites

Offline edge inference keeps automation running in remote facilities and low-connectivity environments where cloud AI breaks down.

Current Stage

Pilot — and proven on real hardware.

We've built and tested the core EdgeMind runtime in real-world hardware environments, validating the technical feasibility of the architecture across multiple scenarios.

Idea
Prototype
Pilot
Growth
Scaling
  • Working runtime on real hardwareEarly versions of the local-first runtime connect AI agents to embedded devices, sensors and hardware control systems.
  • Validated low-latency local executionTested integrations with Arduino-based systems, edge AI components and custom hardware workflows confirm modular device interaction.
  • Natural-language device controlExperimental workflows let AI interpret instructions and trigger actions across connected devices.
  • Continuous community feedbackInput from students, makers and small-scale hardware communities shaped our focus on accessibility, modularity and low deployment complexity.
Local-first
Runtime architecture
0 cloud
Offline by design
Multi-device
Pilot scenarios validated
NL → action
Control workflows tested
The Next AI Era

Not a smarter chatbot. A persistent presence in the real world.

The next wave of AI won't live in a chat window — it will live among us, operating physical systems reliably over the long term. EdgeMind Core is building the infrastructure layer that makes Persistent Physical AI possible.

Why FII Innovators Pitch

Physical AI can't be proven in a simulator.

EdgeMind Core tackles a challenge that needs more than technical innovation — it needs strong global partnerships across industry, infrastructure and emerging AI ecosystems.

Physical AI systems can't be validated by software alone. They demand real-world operational settings, hardware integration partners, and feedback from organizations deploying automation in complex environments.

FII's international network, investment ecosystem and focus on AI-driven societal impact make it uniquely aligned with our mission — to scale EdgeMind from pilot implementations into globally deployable Physical AI infrastructure.

01

Real-world deployment environments

Access to large-scale operational settings and strategic industry collaborators to validate Physical AI where it actually matters.

02

Edge infrastructure scaling

Guidance on scaling edge AI across diverse hardware ecosystems while staying accessible to organizations with limited engineering resources.

03

Mentorship & expertise

Strategic mentorship from global industry leaders, robotics experts and infrastructure-focused investors.

04

Cross-sector partnerships

Connections to the industries and communities that will benefit most from accessible, real-world AI automation.

The Team

Builders from NTHU.

A founding team combining product vision and deep systems engineering, building out of National Tsing Hua University, Taiwan.

HC
Co-Founder · CEO

Hector Chiu

Leads product vision and strategy for EdgeMind Core — translating the Physical AI thesis into a runtime the world can deploy.

YC
Co-Founder · CTO

Yuyi Chang

Architects the local-first runtime, capability layer and edge execution that lets AI agents safely control real-world hardware.

⌁ Originating from National Tsing Hua University · Hsinchu, Taiwan

Let's build the execution layer

Give AI a body in the real world.

We're connecting EdgeMind Core with the partners, deployment environments and investors who will help bring Persistent Physical AI to life.