Skip to content
Download for Mac

Web Search

Type @web-search followed by your query to pull live web results directly into the conversation. QARK supports several search backends — from a free local browser to dedicated search APIs — and tries them in priority order until one returns results.

Web search call blocks in a conversation

Cost: Free. No API key required.

QARK launches a locally installed Chromium-based browser in headless mode to perform real searches. It auto-detects the first available browser on your system: Chrome, Brave, Edge, or Chromium. You can override this with a custom browser path in Settings.

Search engines: DuckDuckGo (default) or Bing (fallback). DuckDuckGo occasionally serves CAPTCHAs for automated requests — QARK detects this and falls back to Bing automatically.

How it works: The browser navigates to the search engine with your query, renders the full page including JavaScript, extracts and parses results, and returns them to the conversation. Each search has a 20-second timeout before falling through to the next provider.

Several LLM providers offer native web search injected directly into their API calls. The model searches the web as part of its reasoning — no separate tool call required.

ProviderApproximate CostNotes
Anthropic~$10 per 1,000 searchesClaude models with web search capability
OpenAI~$30 per 1,000 searchesGPT models with browsing
Gemini~$14 per 1,000 searchesGoogle’s Gemini with grounding
Perplexity SonarIncluded in token costSearch is inherent to Sonar models
OpenRouter (Exa)~$4 per 1,000 searchesExa plugin via OpenRouter

Toggle per provider in Settings → Providers. Provider built-in search only works when no other tools are active in the conversation — provider search injection conflicts with standard tool-use formatting.

Dedicated search services that return structured results. Each requires its own API key configured in Settings.

ProviderFree TierNotes
Brave SearchLimited free tierPrivacy-focused, independent index
ExaLimited free tierNeural search, semantic understanding
TavilyLimited free tierOptimized for AI agents
PerplexityPaidAnswer engine with citations
OpenAIPaidWeb search via OpenAI API
OllamaFree tier availableLocal search capability
JinaLimited free tierReader-optimized search
Parallel AIPaidMulti-source aggregation
ValyuPaidEnterprise search

All enabled search providers appear in a drag-to-reorder list in Settings. QARK tries them top to bottom:

  1. First provider in the list attempts the search
  2. If it fails (timeout, error, CAPTCHA, empty results), QARK falls through to the next provider
  3. First successful result wins — subsequent providers are skipped
  4. The process continues until a provider returns results or all providers are exhausted

Reorder the list to match your preferences: free options first to minimize cost, or premium APIs first for quality.

Web search settings

The Settings UI displays stats for each search provider:

  • Enable/disable toggle — activate or deactivate individual providers
  • API key field — securely store credentials
  • Searches this month — running count of searches made
  • Free tier remaining — for providers with usage caps
  • Estimated cost — calculated from search count and provider pricing