Getting started

Sirius needs a model before the chat can answer. Everything else — the editor, extensions from Open VSX, your settings — works without one.

Install and open Sirius. The install guide covers every platform. On first launch a toast says ★ Sirius AI — multi-model chat with thinking, tools and agent mode. Add a provider key, or start Ollama and Sirius finds your models. with two buttons, Set API Key and Select Model. It shows once. If the editor’s own welcome overlay appears first, it does not set up a model; close it and carry on.
Give it a model — one of two ways.

A provider key. Open the Command Palette (Ctrl+Shift+P) and run Sirius: Set API Key. Pick the provider — Anthropic Claude, Google Gemini, OpenAI, OpenRouter, Groq, DeepSeek, Mistral or xAI Grok — and paste the key into the password field. The prompt says where it goes: stored in the system keyring — never written to settings.json. A toast confirms ✅ Anthropic Claude key saved to the system keyring, and the provider’s models appear at once; no reload. Sirius does not test the key when you save it — the first message does.

Or a local model. Install Ollama, then:
Terminal window
```
ollama serve          # if it is not already running as a service
ollama pull qwen3     # any model you like
```
Sirius looks for Ollama at http://localhost:11434 and lists every model it finds — nothing to configure, no key, and nothing leaves your machine. See Local models for LM Studio, llama.cpp and vLLM.
Pick the model. The Command Palette’s Sirius: Select AI Model: every provider you have configured, with 🧠 for models that can reason and 👁️ for models that accept images, then a detected group for each local server. The choice becomes the default (sirius.ai.defaultProvider and sirius.ai.defaultModel), the status bar shows ★ model name, and clicking that status item reopens the picker. The chat view has its own model picker too, grouped by provider, for a per-conversation choice.
Say what you want. Open the chat with Ctrl+Shift+I (Sirius: Open AI Chat; the editor’s own Ctrl+Alt+I works as well). Sirius is the default participant, so you type into the box that says Describe what to build and press Enter. Switch between Ask, Edit and Agent in the chat input — see Chat, edit and agent.

If nothing answers

With neither a key nor a running Ollama, the chat replies: No model is available yet. Add a provider key with Sirius: Set API Key (Ctrl+Shift+P), or start Ollama and pull a model — Sirius finds it automatically. A provider without a key is not listed in the picker at all. More cases are in Troubleshooting.

Thinking

Models that can reason (the Claude and Gemini models marked 🧠) reason by default. Sirius: Toggle Thinking Mode turns it off and on for the session’s default; Sirius: Set Thinking Effort Level picks Low, Medium, High, Extra High or Max. The status bar shows 🧠 beside the model name while thinking is on for a model that supports it. Local and OpenAI-compatible models are not sent a thinking parameter; if their server returns reasoning text, Sirius shows it.

Defaults worth knowing

Setting	Default
`sirius.ai.defaultProvider`	`anthropic`
`sirius.ai.defaultModel`	`claude-opus-5`
`sirius.ai.thinking.enabled`	`true`
`sirius.ai.thinking.effort`	`high`
`sirius.ai.maxTokens`	`16384`
`sirius.ai.temperature`	`0.7`

The full list is on the Settings page, generated from the shipped manifest.