Deployable
One signed artifact format, per-SoC execution profiles, verified on device before it runs. The same discipline OS images already get.
About
Product, not projects.
ByteHub AI LLC is a California deep-tech startup building the model lifecycle and policy layer between AI capability and real-world device operations. Deploying AI to a device fleet should not require a one-off integration project — it should be product infrastructure with firmware-grade release discipline.
Mission
Make edge AI deployable, governable, and safe to operate.
01
02
Governable
Policy decides what runs locally, what data may leave, when a cloud model may be called, and which actions need approval. Every crossing is audited.
03
Safe to operate
Staged rollout, observability, and a tested rollback path. Offline is the normal case, not an error.
Founder
Built by an engineer who has run the layers AI products ignore.
Jerry — Founder / CEO. Embedded systems engineer. NXP i.MX95 bring-up, Yocto, Linux kernel, U-Boot, and BSP development. RAUC firmware updates, partitioning, secure release workflows, and device debugging. DDS, messaging, fleet and device communications, and distributed systems. Edge AI integration, model deployment, and hardware-aware optimization. Embedded systems work in energy-storage and real-world device environments.
ByteHub's proof standard follows from that background: measured on our own bench, then validated on a customer workflow. Nothing on this site is called measured until it is.
Low-level systems
Bootloader, kernel drivers, Device Tree, Yocto / BitBake BSP layers for physical target silicon.
Secure OTA
Atomic, signed update frameworks (RAUC, SWUpdate) applied to model releases.
Hardware bring-up
i.MX95-class, Jetson Orin, and Jetson Thor boards on the bench; schematic to first boot to reproducible image.
Fleet systems
DDS and messaging for device fleets; telemetry, staged rollout, rollback.
Stage
Pre-seed, self-funded, raising.
Pre-prototype today. i.MX95-class, Jetson Orin, and Jetson Thor hardware are in hand; quantized inference validation is underway; the EdgeRuntime, signed-artifact, policy-gate, and rollback architecture is defined. We are engaging the NVIDIA, Google Cloud, AWS, and Microsoft startup programs and selecting 2–3 industrial Linux-device design partners.
Design partners
Industrial OEMs, fleet operators, and robotics / inspection integrators who will evaluate the runtime and control plane on real hardware.
Design-partner inquiry →Investors
Pre-seed SAFE to convert technical advantage into design-partner proof and seed-ready evidence.
Investors & partners →Ecosystem programs
Silicon-vendor and cloud startup programs for technical guidance, credits, and introductions to OEMs and integrators.
Cloud partner program →
Collaborate
Work with us.
Design partners, investors, and cloud or silicon programs: one address, short reply.