Settings

All of Sirius’s own settings live under sirius.ai in File → Preferences → Settings (search for Sirius), or in settings.json. This table is generated from the extension manifest that ships in the build, so it cannot drift from the product — the descriptions below are the ones the Settings editor shows.

API keys are not settings. They live in the system keyring and are set with the Sirius: Set API Key command; see Providers and keys.

Current settings

Setting	Default	What it does
sirius.ai.defaultProvider string: anthropic · gemini · ollama · openai · openrouter · groq · deepseek · mistral · xai · lmstudio · llamacpp · custom	anthropic	Default AI model provider
sirius.ai.defaultModel string	claude-opus-5	Model used for AI chat. Prefer the "Sirius: Select AI Model" command, which lists everything the configured providers offer.
sirius.ai.ollama.endpoint string	http://localhost:11434	Ollama server endpoint
sirius.ai.thinking.enabled boolean	true	Enable thinking/reasoning mode for supported models (Claude Opus, Sonnet, Gemini 3.5)
sirius.ai.thinking.effort string: low · medium · high · xhigh · max	high	How deeply the AI should reason before responding
sirius.ai.maxTokens number	16384	Maximum tokens in AI response
sirius.ai.temperature number	0.7	AI response creativity (0 = deterministic, 2 = very creative)
sirius.ai.streamResponses boolean	true	Stream AI responses token by token
sirius.ai.enable object	{"*":false}	Tab completion, per language. "*" is the default for every language; a language id such as "markdown" or "json" overrides it. This is the switch the chat status dashboard toggles. Completions need a local FIM model — Ollama or llama.cpp; see `sirius.ai.completions.model
sirius.ai.openai.baseUrl string	https://api.openai.com/v1	Override the OpenAI endpoint — e.g. an Azure deployment or a corporate proxy.
sirius.ai.lmstudio.baseUrl string	http://localhost:1234/v1	LM Studio local server endpoint.
sirius.ai.llamacpp.baseUrl string	http://localhost:8080/v1	llama.cpp or vLLM OpenAI-compatible endpoint.
sirius.ai.custom.baseUrl string	""	Base URL of any other OpenAI-compatible service, including the /v1 suffix.
sirius.ai.completions.model string	auto	Model for Tab completion: "auto" picks a running Ollama code model automatically; or "ollama/<model>"; or "llamacpp" for the configured llama.cpp server.
sirius.ai.completions.maxLines number	12	Longest suggestion Tab completion will offer, in lines.
sirius.ai.ollama.largeModelBytes number	12000000000	Refuse to load Ollama models larger than this many bytes — protection against freezing the machine with a model that far exceeds its memory. 0 disables the check.
sirius.ai.nextEditSuggestions.enabled boolean	false	Predict your next edit after each change and offer it as a Tab-able diff (experimental; needs a local FIM model). Requires Tab completion to be on — see sirius.ai.enable.

Deprecated settings

These still work for one more release cycle but should not be used in new configurations. The replacement is named in each row.

Setting	Default	What it does
sirius.ai.gemini.apiKey string	""	Google Gemini API key (get one from ai.google.dev) Deprecated: keys are now stored in the system keyring. Run "Sirius: Set API Key" instead.
sirius.ai.anthropic.apiKey string	""	Anthropic API key (get one from console.anthropic.com) Deprecated: keys are now stored in the system keyring. Run "Sirius: Set API Key" instead.
sirius.ai.openai.apiKey string	""	OpenAI API key (get one from platform.openai.com) Deprecated: keys are now stored in the system keyring. Run "Sirius: Set API Key" instead.
sirius.ai.inlineCompletions boolean	false	Enable AI-powered inline code completions (experimental) Use sirius.ai.enable — it is per language and is what the chat status dashboard toggles. true here still turns completions on everywhere.

Manifest version 2.0.0.