Overwatch Node · Local AI Server
Your data should remain under your control. Your AI should run on hardware you own. The Node runs models, agents, memory, RAG, tool calling, and automation on a machine that answers to you.
Local when you want it. Cloud when you need it. Control always.
AMD Ryzen AI 9 HX 370 · 64 GB LPDDR5X · 2 TB NVMe
$999 once
Data sovereignty
Data sovereignty is not a feature toggle. It is the sum of the decisions a system lets you make. On Overwatch Node, you decide:
One DC power input, one Ethernet cable or Wi-Fi 7. That's the whole install.
A local web control center, reachable from any browser on your subnet. No account, no activation server.
Pull open-weight models, wire up agents, or build RAG pipelines from the first-run prompt.
Ask questions of your own documents without them ever leaving the building. Run open-weight models and multi-agent workflows around the clock with nothing metering your usage. Stay in control of your data and your compute. None of it has to leave your network.
If you want a managed cloud chatbot, this isn't it. The Node is a box you own and control, not a hosted service.
Most tools that support a configurable OpenAI-compatible endpoint can connect to the Node by changing the base URL.
export OPENAI_BASE_URL=http://node.local:8080/v1
export OPENAI_API_KEY=local # nothing leaves your networkDesigned for local quantized-model inference and simultaneous AI services, limited by the hardware, not a rate cap or a seat license.
Dedicated acceleration for supported AI workloads, working alongside the CPU and Radeon 890M graphics.
2× USB4 Type-C, HDMI 2.1, 2.5GbE Ethernet, and Wi-Fi 7 for storage, capture, and network workloads.
The Node runs, remembers, and improves in place. Memory and RAG live on the box, so nothing depends on a third party staying in business.
Closed-loop AI
Overwatch Node can operate as a closed-loop local AI environment. Documents, prompts, model responses, embeddings, persistent memory and agent activity can remain on the device during a local session without being sent to an outside AI provider.
When maximum isolation is required, you can disable external AI connections and keep the workflow contained within infrastructure you control.
Three operating modes. One owner.
Local Mode
Models, documents, memory and workflows run locally on Overwatch Node.
Closed-Loop Mode
The session is restricted to the local environment, with external AI services and outbound integrations disabled by the user.
Connected Mode
Deliberately connect approved cloud models, APIs or outside services when additional capability is needed.
Supported local workloads can remain on the device. The degree of isolation depends on the configuration you select.
Data sovereignty is more than a privacy policy. It means controlling the infrastructure where your data is stored and processed.
A vendor promise is not ownership. A cloud account is not sovereignty. Overwatch Node puts the hardware, models, memory and data under your control.
Private local AI processing on a self-hosted AI server you own outright. The compute, the storage, and the network exposure are yours to configure.
An open-source software core with no mandatory subscription and no permanent vendor lock-in. The software on your Node keeps running whether or not any vendor does.
Run the open-weight models you choose. Local document intelligence, private RAG, and persistent local memory live on the device, not in someone else's tenant.
Local AI agents and workflows by default. Optional cloud connectivity only when you deliberately enable it, and only to the services you approve.
90% of organizations see local data storage as inherently safer.
Every unit is built to order, securely imaged, and burn-in tested before it ships.
Reserve your Node · $999| Processor | AMD Ryzen AI 9 HX 370 · 12 cores / 24 threads, up to 5.1 GHz |
|---|---|
| AI accelerator | AMD XDNA 2 NPU, up to 50 TOPS · dedicated acceleration for supported AI workloads |
| Graphics | AMD Radeon 890M |
| Memory | 64 GB LPDDR5X |
| Storage | 2 TB PCIe 4.0 NVMe SSD |
| Networking | 2.5GbE Ethernet, Wi-Fi 7, Bluetooth 5.4 |
| Rear ports | Vertically arranged and recessed into one rear face: 2× USB4 Type-C, 2× USB-A 3.2, HDMI 2.1, 2.5GbE Ethernet, DC power input. The other five faces stay clean. |
| Security | TPM 2.0, Secure Boot, encrypted storage |
| Cooling | Bottom-mounted 92 mm PWM intake fan (target) pulling cool air up through the intake gap. Copper cold plate, vapor chamber or heat-pipe assembly, aluminum thermal frame, NVMe heat spreader. Vertical airflow through the aluminum core, exhausting through discreet upper-rear vents. Four rubber feet raise the enclosure approx. 6-8 mm for intake air. |
| Power | External 120 W power adapter with headroom for the Ryzen AI 9 HX 370, NVMe storage, wireless networking, USB devices, and sustained AI workloads. Exact idle and full-load wattage will be published once measured on production hardware. |
| Dimensions (target) | Height 225 mm / 8.9 in · width across flat sides 105 mm / 4.1 in · max width corner to corner approx. 121 mm / 4.8 in |
| Desk footprint | Under 5 × 5 inches |
| Weight (target) | Approx. 1.4-1.6 kg / 3.1-3.5 lb |
| Enclosure | Premium injection-molded ABS in an ultra-matte near-black finish. Tall, narrow six-sided hexagonal tower built around an internal aluminum thermal chassis and heat spreader. |
| Front face | Clean. No logo, no ports, no buttons, no RGB. |
| Status light | One thin 360° activity light around the bottom edge only. Cold blue by default. Can be completely disabled in OverwatchOS settings. |
| Operating system | OverwatchOS: open-source core, local AI, closed loop, secure layer |
8.9" tall 4.1" wide Under 5" desk footprint 64 GB 2 TB
Height, width, weight, fan size, and final power characteristics are production design targets until a physical engineering prototype has been built and thermally tested.
At just under 9 inches tall with a desk footprint under 5 × 5 inches, Overwatch Node is designed to deliver serious local AI capability without taking over your workspace.
Its vertical hexagonal architecture creates additional internal cooling volume while keeping the desk footprint under 5 × 5 inches.
A quiet 92 mm PWM fan draws cool air through the bottom of the Node and pushes it vertically across the processor, memory, storage, and power components before exhausting warm air through discreet upper vents.
The Node sits slightly elevated on vibration-isolating rubber feet to maintain unrestricted airflow.
A bare machine and a weekend of setup gets you a model runner. Every Node arrives pre-configured with OverwatchOS: models, agents, memory, RAG, workflows. One closed loop running locally, from the first boot.
Interface preview. Illustrative values.
Could you buy bare hardware and install Ollama yourself? Sure. You would have a model runner. The Node ships as a complete system: models, agents, memory, RAG, and workflows already wired into one closed loop, secured, tested, and ready the moment it touches your network.
And the software is open source. Inspect the code, swap models, rewrite agents, and extend workflows however you want. Nothing is locked, nothing is licensed per seat. The machine is yours, and so is the creative freedom.
Overwatch Node does not have to replace your existing servers, NAS, GPU systems, home automation, or other machines. OverwatchOS can connect to and orchestrate the infrastructure you already own, using the Node as the local AI control plane for agents, memory, RAG, workflows, tool calling, and automation.
Run fully local when privacy, control, or offline operation matters. Or optionally connect services such as OpenAI, Anthropic, Google, or other supported providers when you want additional cloud capability. The choice stays with you.
| Overwatch Node | Cloud AI APIs | DIY build | |
|---|---|---|---|
| Cost | $999, one time, free shipping | Metered per token, forever | Parts, plus days of your time |
| Where your data goes | Stays on your network | Leaves your network | Local, once you've configured it |
| Setup | Boot into the control center | API keys per provider | Driver setup, package conflicts, thermal tuning |
| Recurring fees | None | Grows with usage | None |
| Software stack | Pre-integrated, open source | Per-vendor SDKs | You assemble and maintain it |
If you enjoy the build, DIY is a fine path. The Node is for people who want the outcome: the hardware and the open-source stack in one burn-in-tested box.
64 GB of LPDDR5X and the Ryzen AI 9 HX 370 provide the compute and memory for compatible quantized open-weight language models, vision workloads, embedding pipelines, speech engines, agents, and self-hosted AI services. The AMD XDNA 2 NPU provides up to 50 TOPS of dedicated acceleration for supported AI workloads. OverwatchOS includes agent orchestration, document RAG, workflow automation, and local model management.
No. Core operation is local. In Closed-Loop Mode the session is restricted to the local environment, with external AI services and outbound integrations disabled by the user, while local models, local memory, local agents, and RAG stay operational. In Connected Mode you can deliberately connect OpenAI, Anthropic, Google, Mistral, or compatible APIs with your own credentials. There is no subscription and no usage metering.
Yes. The Node doubles as a local LLM server, exposing an OpenAI-compatible API on your network. Most tools, scripts, and libraries that support a configurable OpenAI-compatible endpoint can connect by changing the base URL to the Node's local address.
100. Each unit carries a numbered serial from 001 to 100, and this is the only batch that will ever carry them. Production begins when all 100 slots are reserved. Shipping is free.
Each Node is built to order, securely imaged with OverwatchOS, and burn-in tested. Shipping is free, with estimated delivery 13-14 weeks after production begins.
A desktop gives you bare hardware. You still spend days on driver setup, package conflicts, and stack assembly. The Node boots directly into a local web control center with the local AI stack pre-integrated, on hardware sized for the job: 64 GB LPDDR5X, an NPU providing up to 50 TOPS of dedicated acceleration for supported AI workloads, and quiet bottom-up cooling.
First production run: 100 numbered units
Own the box. Own the loop. Own the data. One payment for a private AI server that answers to you and nobody else. Serials 001 to 100, the only batch that will ever carry them.
Local when you want it. Cloud when you need it. Control always.
$999 once
0 / 100 reserved
Reserve your NodeFrom the blog
