Overview

Msty Nexus gives you ownership of the inference layer behind your AI tools. It sits between approved applications and the models they use, turning local runtimes and online providers into one governed gateway.

Instead of configuring every tool separately, you decide where models run, which credentials they can use, how requests are routed, and what your organization can observe—all from Nexus.

One gateway between apps, models, and credentials

Nexus standardizes the connection between your AI tools and model infrastructure:

• Control AI access from one place. Approve applications, scope client tokens, manage provider credentials, and choose which models are available.
• Use local and online AI together. Keep sensitive or routine work on infrastructure you control, then use online providers when their capabilities are needed.
• Route, balance, and scale model work. Send each request to the right model, distribute traffic across healthy capacity, or combine Macs for larger local workloads.
• See how inference is being used. Monitor activity, usage, account capacity, model health, and operational events without exposing prompt or response content in support-safe history.
• Apply guardrails at the gateway. Centralize key protection, scoped access, routing policy, and sensitive-data handling instead of relying on every connected application to implement them correctly.

Your applications keep a familiar OpenAI-compatible or Anthropic-compatible connection. Nexus handles the infrastructure behind it.

What you can build

A governed model catalog

Connect OpenAI, Anthropic, OpenRouter, compatible endpoints, and local runtimes such as Ollama, llama.cpp, and MLX. Nexus discovers what each model can do and prevents requests from reaching unsupported endpoint families.

A private local AI layer

Run models on your own machine, manage their lifecycle, and keep provider keys behind the gateway. Local network access stays off until you deliberately configure it.

A smarter serving layer

• Smart Routes choose the right model for each request behind one stable model name.
• Smart Balancer spreads work across healthy models and machines for steadier throughput.
• Cluster combines Apple Silicon Macs to run distributed MLX workloads that exceed one machine's capacity.

An operational view of AI

Use Usage, Gauge, and Activity to understand traffic, account runway, live work, and system events. Nexus also surfaces model memory, runtime health, budgets, alerts, and recommendations where available.

How Nexus fits together

Nexus keeps inference close to the machines doing the work while giving you one place to manage it:

• Nexus Runtime does the work. It runs on each managed machine, serves the authenticated gateway, starts local runtimes, protects provider credentials, and keeps machine state local.
• Nexus Console gives you control. The same Console experience is available in the desktop app and, for supported editions, on the web. Use it to manage machines, models, routing, access, usage, and team policy from one workspace.
• The desktop app keeps Nexus within reach. It starts and connects to the local Runtime, provides quick status and controls from the menu bar or system tray, opens the full Console, and protects its local credential in the OS keychain.

Connected AI tools authenticate to Runtime with scoped client tokens. They never need direct access to the provider credentials or operational controls behind Nexus.

Find your way around

The documentation follows the same groups and labels as the Nexus Console left rail:

• Workspace: Overview explains the selected workspace, while Fleet shows connected machines and their readiness.
• Serving: Smart Routes, Smart Balancer, and Cluster control how inference work reaches available capacity.
• Library: Providers, Models, Presets, and Local Runtimes define what Nexus can serve.
• Observe: Usage, Gauge, and Activity show consumption, capacity, and live operations.
• Administer: Teams, Client Tokens, and Settings control people, application access, and machine behavior.

Cross-cutting topics such as Setup, Gateway, Security, and Troubleshooting live under Guides.

Start with one useful path

For a first setup:

1. Prepare one local runtime or connect one provider.
2. Add or sync a model.
3. Create a preset when you want a stable model configuration.
4. Create a scoped client token.
5. Connect an application through the Gateway.
6. Confirm real traffic in Activity and Usage.

Once that path is reliable, add Smart Routes, Smart Balancer, Cluster, Gauge, and stronger policy controls as your workloads grow.