Settings Reference

Every configurable option in QARK. Open Settings with Cmd+, (macOS) or Ctrl+, (Windows/Linux).

General

Setting	Default	Description
Theme	System	System, Light, or Dark. System follows the OS preference.
Send behavior	Enter	Enter sends immediately (Shift+Enter for new line). Cmd/Ctrl+Enter requires the modifier, freeing Enter for new lines.
Show thinking blocks	On	Display the model’s chain-of-thought reasoning in expandable blocks.
Expand thinking blocks	Off	Auto-expand thinking blocks as they stream in, instead of starting collapsed.
Auto-generate titles	On	Uses the embedded model to create a descriptive conversation title after the first exchange.
Close to system tray	On	Closing the window minimizes to the system tray instead of quitting.

Providers

Configured per provider. See Providers for details.

Setting	Default	Description
API key	—	Authentication key. AES-256 encrypted locally.
Base URL override	Provider default	Custom endpoint for proxies, enterprise gateways, or local servers.
Enabled	On	Disable to hide a provider’s models without deleting the API key.

Budget

Setting	Default	Description
Per-provider monthly limit	Disabled	Spending cap in USD per provider. QARK stops requests when the limit is reached. Resets monthly.
Enable/disable toggle	Off	Each provider’s budget must be explicitly enabled.

Tools & MCP

Settings are organized into tabs within the Tools & MCP section.

Web Tools

Setting	Default	Description
Web search providers	—	Enable/disable individual search providers (Brave, Exa, Tavily, Perplexity, etc.). Each requires an API key.
Provider priority	Drag order	Drag to reorder search providers. First successful result wins.
Local browser search	Auto-detected	Free tier using a locally installed Chromium-based browser. Search engine: DuckDuckGo or Bing.
Custom browser path	Auto	Override the browser executable path for headless search/fetch.
Provider builtin search	Per-provider toggle	Enable native search capability in Anthropic, OpenAI, Gemini, Perplexity, OpenRouter.

Documents & RAG

Setting	Default	Description
Embedding provider	—	Provider for vector embeddings (Voyage AI, Jina AI, Cohere, OpenAI).
Embedding model	—	Specific model for embeddings. Larger dimensions improve accuracy; smaller are faster.
RAG generation provider	—	Provider for generating answers from retrieved context.
RAG generation model	—	Model for RAG answer generation.
Reranker provider	None	Cross-encoder reranking provider (Cohere, Jina, Voyage). Optional.
Reranker model	—	Specific reranking model.
RAG threshold %	30	Minimum percentage of context window for a document to trigger full RAG vs. direct injection.
Image extraction provider	—	Provider for extracting text from images in PDFs.
Image extraction model	—	Vision model for PDF image extraction.

Image Generation

Setting	Default	Description
Image generation provider	—	Default provider for @image-generation (OpenAI, Gemini, xAI, OpenRouter).
Image generation model	—	Default model (DALL-E 3, Imagen, Flux, etc.).

Video Generation

Setting	Default	Description
Video generation provider	—	Default provider for @video-generation (OpenAI, Google, Runway).
Video generation model	—	Default model (Sora, Veo, Gen-3, etc.).

Unix Commands

Setting	Default	Description
Default working directory	Empty	Starting directory for @unix-commands. Per-conversation overridable.

MCP Servers

Setting	Default	Description
Server list	Empty	Registered MCP servers. See MCP Integration for setup.
Auto-connect	On	Servers connect automatically on app launch.
Per-server tool toggles	All enabled	Enable/disable individual tools discovered from each server.

Embedded Model (Inference)

Setting	Default	Description
Embedded model	Auto	Local or cloud model used for auto-naming conversations and groups. A fast, inexpensive model works best.

Sparks & Flows

Setting	Default	Description
Dismiss on blur	On	Overlay closes when it loses focus. No effect when pinned or streaming.
Default mode	Flows	Which tab opens by default: Spark or Flows.
Default output action	Copy	Action after a Spark/Flow completes: copy, paste, insertBefore, insertAfter, or openInChat.
Default chat agent	—	Agent used for regular conversations.
Default chat provider	—	Provider for regular conversations.
Default chat model	—	Model for regular conversations.
Default Spark agent	—	Agent used for Spark invocations.
Default Spark provider	—	Provider for Spark invocations.
Default Spark model	—	Model for Spark invocations.
Default compaction provider	—	Provider for context compaction/compression.
Default compaction model	—	Model for compaction. If unset, uses the chat model.

Model Visibility

Setting	Default	Description
Disable models	All visible	Hide specific models from the model picker.
Disable model groups	All visible	Hide entire model families (e.g., all GPT-4 variants) from a provider.

Custom Categories

Setting	Default	Description
Prompt category order	Default	Reorder categories in the prompt library. Drag to rearrange.
Flow category order	Default	Reorder categories in the flow library. Drag to rearrange.
Collapsed categories	None	Which categories start collapsed in prompt/flow libraries.