Nova Studio

中文说明

Nova Studio is a Windows desktop workbench for running local LLMs through WSL2.

It is built for the workflow of:

add a model -> create/start an instance -> expose a local API gateway -> connect Trae or other OpenAI-compatible clients -> inspect logs and runtime behavior

This project is not just a thin GUI over one backend. It combines:

• a native desktop shell
• multi-engine instance management
• a local OpenAI / Anthropic compatible gateway
• WSL process orchestration that survives Node.js restarts
• observability and debugging tools for day-to-day local model usage

Download

Windows portable (no install): nova-studio-v0.2.0-windows-portable.zip (~32 MB — unzip and run)

What's included? · Full release notes

Status

v0.2.0 — Nova Studio rebrand + port & cold-start hardening

First public release: v0.1.0-alpha

What is already working well:

• managing local vLLM, SGLang, and llama.cpp instances
• connecting Trae through the local gateway
• using the local model for real coding sessions
• checking logs, health, model lists, and basic runtime metrics

What is still evolving:

• installation smoothness across different Windows / WSL / driver setups
• public release polish
• some advanced features such as system proxy hijacking, embedded finetune tooling, assistant tools, and MCP management

Why This Exists

Most local LLM tools stop at "the model is running".

This project tries to solve the next layer of problems:

• keeping multiple model instances manageable from one desktop UI
• making local models easy to plug into coding tools
• reducing friction when switching between engines
• giving you one place to inspect logs, routing, and request behavior
• handling long IDE-style contexts more deliberately than a bare command line setup

Core Workflow

1. Add a model preset or restore a discovered local model.
2. Create an instance card or launch a new instance directly.
3. Use the built-in gateway as the stable external endpoint.
4. Generate Trae connection settings from a running instance.
5. Watch logs, health, models, and runtime status from the desktop app.

About Preset Templates

The preset templates shown on first launch are starter configurations, not bundled or already-running models.

In a fresh setup, users should see the same preset template list after the app initializes successfully. Actual inference still depends on their own WSL environment, engine installation, model files, and GPU setup.

Screenshots

Main Console

The main workspace combines service cards, instance state, and deployment controls in one desktop view.

Logs and Observability

Logs are available as a first-class workflow instead of an afterthought, which makes it easier to inspect startup issues and runtime behavior.

Runtime Metrics

The app also exposes a dedicated runtime view for quick health and throughput inspection.

Add Model and Deploy Flow

New models can be added and turned into reusable presets or instance cards from the same workflow.

Settings and Integration

Settings bring together integration-related options, including the workflow used to connect external coding tools.

Floating Mode

A compact floating view keeps quick status and controls visible while you work in other apps.

Main Capabilities

Stable core

• multi-engine management for vLLM, SGLang, and llama.cpp
• instance cards for start / stop / restart / inspect flows
• local API gateway with OpenAI-compatible and Anthropic-compatible entry points
• Trae integration helper with copyable connection fields
• request logs, health checks, model listing, and runtime metrics
• disk model discovery and preset import / export

Advanced / still maturing

• system proxy hijacking for redirecting editor traffic to the local gateway
• embedded Unsloth Studio workflow
• built-in assistant panel with tool use
• local skills and MCP server management
• request recording and deeper analytics

Architecture

nova-studio.exe (Go / Wails)
├── React frontend in a native window
├── Node.js API service on 127.0.0.1:3001
├── Go WSL manager on 127.0.0.1:3002
└── model engines running inside WSL2
    ├── vLLM
    ├── SGLang
    └── llama.cpp

The Go side manages WSL-backed processes directly so model instances do not depend on the Node.js API process staying alive.

Repository Layout

vllm_5090D/
├── vllm-launcher/      # Go/Wails desktop shell + frontend
├── launcher/           # Node.js API, gateway, orchestration, routes
├── finetune/           # finetune helpers and data factory scripts
├── BUG_ANALYSIS.md
└── SECURITY_AUDIT.md

Requirements

• Windows 10/11
• WSL2
• NVIDIA GPU for practical local inference
• Node.js 20+
• Go 1.25+
• Wails CLI for source builds

Engine-specific runtime dependencies are not bundled. You are expected to prepare your chosen WSL environment for vLLM, SGLang, or llama.cpp.

First-Time Setup

For first-time Windows users, especially on machines that do not already have WSL2 installed, follow the step-by-step guide here:

• Windows Quick Start
• Portable Release Quick Start

If you want a fast machine check before opening the app:

powershell -ExecutionPolicy Bypass -File .\scripts\windows-preflight.ps1

If your machine is already ready and you now need the WSL-side engine setup:

• WSL Engine Setup

If you only want to distribute a prebuilt package so users can quickly verify that the app opens correctly on their machine, see the portable release notes:

• Portable Release Quick Start

Development

Install dependencies:

cd launcher
npm install

cd ../vllm-launcher/frontend
npm install

Run in development:

# Terminal A
cd launcher
npm run server:dev

# Terminal B
cd vllm-launcher/frontend
npm run dev

# Terminal C
cd vllm-launcher
wails dev

Build a Release Directory

cd vllm-launcher
.\build-release.ps1

This assembles a local release folder containing:

• nova-studio.exe
• server.cjs
• launcher-data/

The project currently targets a Windows desktop release flow, not a cross-platform package.

To produce a shareable portable .zip for GitHub Releases:

powershell -ExecutionPolicy Bypass -File .\scripts\package-portable-release.ps1

That script builds a versioned folder and a matching zip such as:

• release-artifacts/nova-studio-vX.Y.Z-windows-portable/
• release-artifacts/nova-studio-vX.Y.Z-windows-portable.zip

For the intended user-facing behavior and limits of that package, see Portable Release Quick Start.

Security Notes

• runtime data lives under launcher-data/
• local logs, recordings, proxy certificates, and keys should never be committed
• review SECURITY_AUDIT.md before public redistribution

If you plan to publish binaries, double-check:

• API key handling
• request recording defaults
• proxy / CA certificate behavior
• WSL command execution paths

Limitations

• Windows + WSL2 is the primary target environment
• the project is still being cleaned up for first public release
• some tooling under finetune/ is experimental and less polished than the main app

Feedback

This is the first public release of the project. Issues and real-world setup feedback are especially helpful right now.