Getting started
Your first ten minutes with Nyx AI.
The actual first-run order, with advanced detail left in the complete reference.
1. Check the basics
You need Windows 10 or 11 on x64 hardware and must be 18 or over. Nyx AI includes no model or inference runtime. For local AI, install Ollama separately and pull a model; for cloud AI, bring an account or key. Provider charges and terms may apply.
2. Install from the official Store listing
Use the Microsoft Store route linked from Installation details. The website does not distribute an installer. Read and accept the English EULA and Terms in the app — the same gate confirms the Privacy Notice was provided to you — or decline and quit.
3. Choose a folder or Chat
After consent, the first screen asks you to open a project folder or continue without a folder in Chat mode. Chat can answer questions and work with attachments. A folder opens Nyx AI Lab and enables workspace tools.
4. Connect a model
The setup wizard checks Ollama and lets you configure provider connections and optional toolchains. A loopback Ollama model can stay on this PC. Remote Ollama, any model ending :cloud or -cloud, and direct cloud providers are off-device. Model reference.
5. Keep the safety defaults
Start with Automatic isolation and Default approvals. Automatic tries AppContainer and can fall back to Standard; Standard is not a filesystem jail, and at start-up or where neither backend can be created a screened command can run with no OS jail — the tier badge shows what actually ran. Choosing an explicit Sandbox level fails closed if AppContainer is unavailable. Isolation details.
6. Confirm how you want to work
The wizard’s final screen confirms Chat or Nyx AI Lab. If you already chose Chat on the welcome screen, the Chat card is preselected; select it to finish. You can switch modes later. Mode reference.
7. Try one small task
Ask for a summary of an attached document in Chat, or ask Nyx AI Lab to explain one file in your project. Read every approval, inspect edits and test the result. AI output can be wrong or insecure even when the tool call itself was contained.
Need more detail?
The complete reference covers approvals, shell tools, documents, previews, privacy and updates. The Security page explains the boundaries without setup instructions getting in the way.