furcate

About

We build the substrate edge AI runs on.

Coast watches the ocean. Grid watches the grid. Industry watches the floor. Furcate is the layer that lets all of them — and any other intelligence platform — run on the customer’s hardware, in the customer’s network, under the customer’s key.

Vision

The cloud was the start. The edge is where it lives.

For a decade, AI meant uploading data to a hyperscaler and waiting for a response. In 2026 that’s no longer the default. Privacy laws, network economics, latency requirements, and sovereignty concerns mean inference runs at the edge — on a Pi, on a Jetson, on an ESP32 in a sensor pod, on an air-gapped gateway in a vault. The substrate that supports this shift is what Furcate is.

We don’t train novel foundation models. We compose the best open-source projects in edge AI — runtimes, orchestrators, federated-learning frameworks, mesh networks, hardware abstractions — into one coherent fabric. Cited end-to-end. Sovereign by default.

Principles

What we build, and why.

Sovereign by design

Inference at the customer's perimeter. EU-residency, on-premise, or air-gapped — the customer chooses. GDPR, HIPAA, FIPS, NIST, CMMC compliant. Data leaves the trust boundary only with explicit, logged consent.

Open foundations, never black boxes

TensorRT Edge-LLM, LiteRT, ONNX Runtime, ExecuTorch, OpenVINO, NVIDIA FLARE, Flower, KubeEdge, OpenYurt, K3s, Akri, WasmEdge, Wasmtime, Dapr, EdgeX Foundry — all cited, all composable.

Hardware-agnostic

Raspberry Pi, NVIDIA Jetson family, Hailo, Coral, Arduino, ESP32, RP2350, ARM Cortex-A and -M, x86 — and bring-your-own custom silicon. The runtime is the abstraction; the hardware is your choice.

Composable, not opinionated

Furcate doesn't tell you which runtime to use. It composes the right one for the job — Wasm for cold-start, TensorRT for Jetson LLMs, LiteRT for microcontrollers, OpenVLA for cobots — and gives you one operations layer over all of them.

Federated-first

Raw data shouldn't leave the device. NVIDIA FLARE + Flower + ExecuTorch make that practical at production scale — Eli Lilly TuneLab, Taiwan MOHW, Tri-Labs (Sandia/LANL/LLNL) all use this pattern.

Auditable autonomy

Every model swap, every OTA campaign, every federated round is logged with cryptographic chaining. Replay viewer for operators; structured event log for auditors. Designed for IEC 62443, NIS2, and OEM customer audit regimes.

The Ipnops family

Four platforms, one philosophy.

  • Ipnops Coast

    Coastal intelligence

    shorelines · reefs · mangroves · sea level

  • Ipnops Grid

    Energy & physical systems

    residence · CRE · utility · cities · critical energy infra

  • Ipnops Industry

    Industrial automation

    manufacturing · process · heavy machinery · agriculture · logistics

  • Furcate

    Sovereign edge AI fabric

    runtime · network · compute · the layer the others sit on

Press & careers

Talk to us.

Press inquiries at press@ipnops.com. Engineering and research candidates: tell us what edge fleet you’ve actually shipped to. eng@tenzro.com.