Platform

Platform

AI Sandboxing. Essential for in-house AI.

A new open model every week. Every product team wants one embedded yesterday. How do you evaluate them, run them safely, and defend a sovereign AI claim? nFOX answers with a resident runtime on your hosts and one console that drives it.

Set up a demo

nFOX runs on your GPU hosts, contains every model it serves in the Blockhouse sandbox, and keeps runtime evidence inside your sovereign envelope.

Scan Surface

Install

Stand Up a Model Easily.

Any host Cloud VM, rented GPU box, or a machine in your rack: any SSH-reachable Linux host with an NVIDIA GPU.
Self-preparing The nFOX warmup process checks the model and automatically prepares everything it needs to run.
Fully hydrated nFOX always verifies model provenance and integrity before running.

Evaluate

Every model walks the same path.

Scan Surface checked before weights land: config, tokenizer, code-execution flags. Blocking findings stop the pipeline.
Attest Per-file SHA-256 seal on your hardware. From then on, the model is its hash.
Stress test Adversarial prompts against the live, contained model, ours and yours.

Contain

Models run inside the Blockhouse sandbox. Not beside it.

Kernel-enforced Containment below the application layer, not a wrapper.
One door No inbound access; a kernel-policed, audited porthole is the only way in or out.
Inside cannot lift it The model process cannot lift the boundary from inside.
Drift halt SHA-256 constant attestation enforced while live: serving halted on the spot.
Blockhouse technical specs

Serve

One box. Multiple models.

Engines Choose the engine that works for you: vLLM, SGLang, or llama.cpp. Model parameters are exposed in an intuitive UI.
Quantization Choose from multiple off-the-shelf quantization modes.
Model access Every production deployment has an OpenAI-compatible API endpoint. Connect it to the tools you already use.
Fleet Ten models or one: a production table, per-model connect, reconfigure, stop, and resend.

Observe

AI oversight that fits your security stack.

Continual integrity Checks keep running after startup, across model and enforcement state.
Audit trail Every request counted, every halt recorded. A halted action is also evidence.
Endpoint Detection and Response Make AI hosts part of your EDR strategy, with endpoint monitoring alongside Blockhouse workload containment.
SIEM visibility Bring AI audit evidence into your SIEM workflow to review policy violations and containment events alongside your other security signals.

Control

One console. Yours alone.

Sovereign envelope Prompts, vectors, tokens, tool calls, runtime services, and audit paths stay inside the enterprise control framework.
Single-tenant No other company's models, users, or scans share your nFOX environment.
Outbound-only Your GPU hosts dial out; zero inbound ports opened.
Inspectable evidence Canonical JSON and Parquet you can inspect, not an opaque database.

Next Step

Review the platform against your deployment requirements.

Contact nFOX