Enterprise AI inference,
running on-premise
NeuronCluster lets you run any model on infrastructure you control. Manage fleets of compute nodes through distributed gateways and one central hub - so sensitive workloads stay inside your walls, with no per-token cloud bill.
- Runs on your hardware
- Data never leaves your network
- Any model, one control plane
- 100%
- On your infrastructure
- 0
- Bytes leaving your network
- Any
- Model, framework or modality
- 1
- Hub to manage every node
Built for teams with strict data-residency requirements
Everyday apps, with AI that runs on your hardware
The NeuronCluster desktop suite brings local AI into the tools you already use. Every app talks to your own Neuron Network — so the intelligence is yours and the data never leaves your walls.
Focus Mail
Email & calendarThe free mail client that answers to your AI, not a mail cloud.
- How local AI helps
- Drafts replies, summarizes long threads and scores spam using models on your own Neuron Network.
- Privacy first
- Mail stays with your IMAP provider and on this machine. Message text only ever goes to the gateway you point it at.
Soma
Database IDEA database IDE that connects and queries the minute it installs.
- How local AI helps
- Writes and explains SQL against your schema, with the assistant served from your own gateway.
- Privacy first
- Only table and column names reach the model — never rows. Passwords are encrypted locally under a key held by the OS keychain.
Neuron Hub
Fleet control planeThe control plane for the machines your apps borrow their AI from.
- How local AI helps
- Provisions gateways and GPU compute nodes over SSH, deploys models and shows live sync and health.
- Privacy first
- No hosted control plane. SSH keys and deployment details stay on your machine; your fleet runs on hardware you own.
Privacy first
Each app works on your machine first. Nothing is sent to a third-party AI provider, and there is no hosted model in the loop.
Powered by Neuron Network
Point any app at your Neuron Network gateway and it uses the models your own GPUs serve, over an OpenAI-compatible API.
Your account, not your data
Signing in backs up app settings — connections, rules, saved queries — and lists your devices. Mail, database rows and passwords never leave your machines.
Download the apps for macOS, Windows and Linux
Create a free NeuronCluster account to get the installers and sign your machines in.
One platform. Total control.
A management hub, a gateway layer and a compute fleet - the building blocks of self-hosted inference, designed to scale with your organization.
Central Management Hub
One control plane for every model, node, and team. Register models, define routing, manage users and roles, and watch live telemetry across your entire fleet.
Distributed Gateways
Stateless gateways load-balance inference over gRPC, HTTP and WebSocket, each serving its own pool of compute nodes for locality and resilience.
Hardened Compute Nodes
Run models on your own GPUs and CPUs inside layered sandboxes - seccomp, namespaces, Landlock and resource limits keep every execution isolated.
Any model, one API
LLMs, vision, audio, embeddings or your own weights. Drop in a model and call it through a single, consistent REST/gRPC interface.
Data stays home
Inference happens entirely within your perimeter. No data egress, no third-party processors, deployable fully air-gapped.
Predictable economics
Saturate hardware you already own. No metered per-token billing, no idle-GPU waste.
A control plane and a compute fleet, cleanly separated
Requests flow from your applications through stateless gateways to sandboxed compute nodes. The hub orchestrates the whole topology - while every byte of data stays inside your network.
Your applications
REST / gRPC / WebSocket clients
Central Management Hub
Models · nodes · routing · users · telemetry
Gateway A
Subnet · region 1
Gateway B
Subnet · region 2
Node 1
GPU
Node 2
GPU
Node 3
CPU
Why teams self-host with NeuronCluster
The same platform serves three needs at once - velocity for builders, efficiency for operators, and control for security teams.
For developers
A single API to every model and modality, running on endpoints you own. Build, test and ship AI features without wrestling vendor limits or rate caps.
- One SDK, any model
- Local, low-latency endpoints
- Streaming & batch
Service optimization
Consolidate scattered GPU spend onto hardware you already run. Pool capacity across nodes, eliminate idle waste and cap costs with predictable economics.
- No per-token billing
- Fleet-wide utilization
- Elastic node scaling
Sensitive workloads
Keep regulated and confidential data inside your walls. Run inference air-gapped, satisfy data-residency mandates and prove it with full audit logging.
- Zero data egress
- Air-gap capable
- Compliance-ready
From zero to private inference in four steps
Most teams stand up their first self-hosted model on day one.
Deploy the hub
Stand up the Central Management Hub in your datacenter, private cloud or VPC. One binary plus a database - no external dependencies required.
Connect gateways & nodes
Attach gateways per region or subnet, then register compute nodes. Models sync down automatically and nodes report health to the hub.
Route any model
Publish models to subnets and define routing policies. Clients call a single endpoint; the platform handles discovery, balancing and isolation.
Operate with confidence
Monitor latency, utilization and audit logs from the hub. Scale by adding nodes - no data ever leaves your perimeter.
Ship AI features against infrastructure you own
One integration for every model and modality. Point your SDK at your own hub and start building - no data leaves the network.
- Unified REST, gRPC and WebSocket APIs
- Drop-in SDKs for the stacks your teams already use
- Streaming responses and batch processing
- Self-hosted endpoints - no third-party in the path
import { NeuronCluster } from "@neuroncluster/sdk";
const nc = new NeuronCluster({
endpoint: "https://hub.internal.acme.com",
apiKey: process.env.NC_API_KEY,
});
// Same call, whether the model is an LLM,
// a vision model, or your own fine-tune.
const res = await nc.inference.create({
model: "llama-3-70b-instruct",
input: { prompt: "Summarize this contract..." },
});
console.log(res.output);Built for your most sensitive workloads
When data cannot leave the building, NeuronCluster keeps inference inside your perimeter without compromising on capability.
Fully self-hosted
Deploy in your datacenter, private cloud or VPC. Run completely air-gapped when required.
No data egress
Prompts, inputs and outputs never leave your network. No external processors, ever.
Sandboxed execution
Every model runs behind seccomp, namespaces, Landlock and strict resource limits.
Role-based access
Granular RBAC over models, nodes and projects, scoped to teams and environments.
Full audit trail
Every request, model change and admin action is logged for compliance review.
Signed results
Compute nodes cryptographically sign outputs so you can verify provenance end to end.
License the platform. Own the infrastructure.
NeuronCluster is licensed per deployment, not metered per token. You bring the hardware; we make it a managed inference fleet.
Starter
Self-hosted
For teams piloting private inference on a single site.
- Central Management Hub
- Up to 2 gateways
- Unlimited compute nodes on your hardware
- REST, gRPC & WebSocket APIs
- Community support
Business
Custom
For organizations scaling inference across teams and regions.
- Everything in Starter
- Multi-region gateways & subnets
- Role-based access control
- Audit logging & observability
- Priority support & onboarding
Enterprise
Custom
For regulated, air-gapped and mission-critical deployments.
- Everything in Business
- Air-gapped & on-prem deployment
- SSO / SCIM & custom RBAC
- Dedicated solutions architect
- Custom SLAs & 24/7 support
Bring inference in-house
See how NeuronCluster runs your models on your infrastructure - with the control, economics and compliance posture your organization needs.