Overview

Msty Nexus is Msty's local-first gateway and control plane for AI models. It gives apps and scripts one stable local API while Nexus manages provider credentials, local runtime lifecycle, model catalogs, presets, and usage visibility.

Start with Setup if this is your first run. If you are connecting another app, go to Gateway. If you are exposing Nexus beyond the local machine, read Security first.

Product model

Msty Nexus has two main parts:

• Msty Nexus Runtime: the background per-machine service that exposes the API, stores local state, starts local runtimes, and protects credentials.
• Msty Nexus app: the menu bar or system tray client that manages Runtime through the public API and stores its local token in the OS keychain.

The app is a client surface, not a separate product. Third-party clients use the same API boundary through local Nexus tokens.

What Nexus does

Nexus combines these capabilities:

• A local OpenAI-compatible and Anthropic-compatible gateway
• Secure provider credential storage with write-only provider keys
• Provider setup for OpenAI, Anthropic, OpenRouter, and compatible endpoints
• Local runtime lifecycle for Ollama, llama.cpp, MLX, and CLI Proxy
• Model discovery, cataloging, install, import, load, unload, delete, sync, health checks, and loaded-memory visibility
• Presets, route packs, Smart Route, thinking controls, prompt caching, and local acceleration controls
• Usage metrics, Gauge account visibility, budgets, alerts, recommendations, and runway checks
• Local Nexus token management, native app pairing, jobs, diagnostics, and audit events

Core capabilities

Nexus includes:

• Gauge as an opt-in AI usage and runway workspace
• Image, speech, transcription, and rerank gateway routes
• Model library health checks and loaded model memory state
• Multipart image edit and audio transcription handling
• Start at Login, Always on Top, and app shell reliability improvements
• Better hidden-model handling in Smart Route and connected clients
• Rich preset generation controls: provider-keyed options, thinking effort/display, Anthropic prompt caching, preset validation, and warnings when a target provider cannot honor a setting

Supported provider and runtime families

Current gateway families include:

• OpenAI-style: models, chat completions, embeddings, Responses, images, audio, and transcription routes where supported by the selected provider.
• Anthropic-style: Messages and token counting routes where supported by the selected provider.
• Rerank: vendor-neutral document reranking through compatible local or OpenAI-compatible runtimes.
• Local runtimes: Ollama, llama.cpp, MLX, and OpenAI-compatible local endpoints.
• Online providers: OpenAI, Anthropic, OpenRouter, and compatible provider records.

Every model advertises the endpoint families it supports. Nexus rejects a request when the selected model is not approved for that route.

App navigation

The first-party app is organized around the operating workflow:

• Overview: runtime health, setup progress, gateway state, catalog counts, and attention items
• Providers: cloud providers, compatible APIs, local runtime discovery, and provider model sync
• Models: model catalog, local model search/install/import, visibility, aliases, GPU setup, and library health
• Usage: local gateway usage, tool counts, route decisions, and acceleration stats
• Gauge: optional AI account limits, reset times, spend, budgets, alerts, and runway
• Settings: runtime connection, network access, client tokens, storage, performance, Start at Login, diagnostics, and notifications

Recommended path

For most users:

1. Prepare one local runtime or connect one cloud provider.
2. Sync or add at least one model.
3. Create one preset for a stable model value.
4. Create a client token.
5. Point another app at the OpenAI-compatible gateway URL.
6. Review usage and diagnostics after real traffic starts.

Keep the first setup intentionally small. Add Gauge, route packs, Smart Route, and preset thinking/cache controls after the gateway path is working.