VLT-001VAULT NODE
SECURESHA256 ✓
37.7749° N122.4194° W
RUNTIME 1.0STABLE
RUNTIME ONLINE CPU 04% MEM 31% LAT 12ms VAULT v1.0 SCROLL 0%
DIFFRACT — BRAND SYSTEM VAULT v1.0 INFRASTRUCTURE THAT CANNOT BE BREACHED EDITION 01 / 06 — MACHINE PRECISION
Runtime v1.0 — Stable Apache 2.0 Brooklyn · NY

Deploy safe, autonomous AI agents in one command.

Enterprise AI Agent Runtime built on kernel-level isolation — Landlock LSM, seccomp BPF, and network namespaces. Multi-provider inference routing with YAML policy control. Infrastructure that cannot be breached.

Scroll
00
AI Models
00
Command Setup
00
Key Leakage
00
Platforms
Landlock LSM seccomp BPF Network Namespaces NVIDIA NIM Anthropic Claude OpenAI GPT k3s / Kubernetes Caddy HTTPS

Built for zero-trust
autonomy.

Six pillars that turn unpredictable agents into auditable infrastructure.

01 / KERNEL

Kernel-Level Isolation

Deny-by-default network egress, API key isolation, L7 TLS MITM inspection, and full privilege separation enforced at the syscall boundary.

02 / INFERENCE

Multi-Provider Routing

Hot-swap between Claude, GPT, Llama, Nemotron and more at runtime — no restarts, no downtime, no leaked credentials.

03 / POLICY

YAML Policy Control

Declarative host, port and method restrictions with shipping presets. Audit-friendly, version-controlled, GitOps-ready.

04 / CHANNELS

20+ Messaging Channels

Native bridges to Telegram, Discord, Slack, WhatsApp and beyond. Deploy a new surface in a single command.

05 / OPS

Production Operations

Watchdog supervision, fork-bomb prevention, integrity diagnostics, and automatic recovery — battle-tested for 24/7 uptime.

06 / HUB

Skills Marketplace

Install vetted agent capabilities from the hub. SHA256-verified, sandboxed by default, removable in one command.

Six layers,
zero trust.

Every request crosses a defined boundary. Every boundary enforces policy. Every policy is auditable from CLI to kernel.

CLI InterfaceNode.js
OpenShell RuntimeRust
Caddy Reverse ProxyL7 TLS
Sandbox EntrypointLandlock
Agent Gateway · NetNSseccomp
Inference RouterMulti-Provider

Infrastructure that
cannot be breached.

Six guarantees baked into the runtime. Not bolted on — enforced at the kernel.

i.

Deny-by-Default Networking

No egress unless declared in policy. Every connection inspected, logged, and rate-limited at the gateway.

ii.

L7 TLS Inspection

Caddy terminates and re-encrypts agent traffic, exposing payloads to policy filters before forwarding.

iii.

API Key Injection

Credentials never enter the agent process. Injected at the proxy layer, scrubbed from every log.

iv.

Operator Approval

Sensitive actions pause for human confirmation via signed channel. Approvals are cryptographically logged.

v.

Fork-Bomb Prevention

Cgroup limits cap process count, memory, and CPU. Runaway agents are killed before they affect the host.

vi.

SHA256 Integrity

Every binary, skill, and policy is hash-verified at load time. Tampered artifacts refuse to run.

Eleven models,
one router.

Switch providers in real time. No SDK reshuffling, no environment swaps, no leaked tokens.

Model
Provider
Switch Command
Status
01
Claude Sonnet 4.5
Anthropic
diffract use claude-sonnet
Active
02
Claude Opus 4
Anthropic
diffract use claude-opus
Active
03
GPT-4.1
OpenAI
diffract use gpt-4.1
Active
04
GPT-4o
OpenAI
diffract use gpt-4o
Active
05
Nemotron 70B
NVIDIA NIM
diffract use nemotron
Active
06
Llama 4 Scout
Meta
diffract use llama-scout
Active

Three steps to
production.

From zero to a sandboxed, multi-provider agent in under five minutes.

i

Install

Single binary. No daemons, no system services, no hidden state.

$ npm i -g diffract
ii

Onboard

Provision the kernel sandbox, register your providers, and spin up the proxy.

$ diffract onboard
iii

Deploy

Hand the runtime a YAML manifest. The runtime handles the rest — securely.

$ diffract deploy agent.yaml

Ready to deploy your
first agent?

Open source. Self-hosted. Audit-ready. Built for teams who refuse to trade safety for speed.