ISO 3166-1 Alpha-2 Country Code (E.g.

AI Gateway gives constructed-in web search capabilities that allow AI models to access present data from the web. This is beneficial once you want up-to-date data that is probably not in the model's coaching information. Seek for all suppliers: Use Perplexity Search, Exa Search, or Parallel Search with any mannequin regardless of supplier. This offers you consistent net search conduct across completely different models. Provider-particular search: Use native net search instruments from Anthropic, OpenAI, Google, or xAI. These tools are optimized for their respective providers and may provide extra features. The perplexitySearch software can be used with any model regardless of the mannequin supplier or creator. This makes it a flexible choice when you want constant internet search conduct across different fashions, or whenever you need to use internet search with a mannequin whose supplier doesn't offer native web search capabilities. To use Perplexity Search, import gateway from ai and pass gateway.tools.perplexitySearch() to the tools parameter. When the mannequin needs present info, it calls the instrument and AI Gateway routes the request to Perplexity's search API.

Perplexity net search requests are charged at $5 per 1,000 requests. See Perplexity's pricing for extra details. Tokens: Total token price range throughout all outcomes. TokensPerPage: Tokens extracted per webpage. ISO 3166-1 alpha-2 country code (e.g., 'US', 'GB') for regional outcomes. Max 20 domains. Cannot mix allowlist and denylist. RecencyFilter: Filter by content recency. Values: 'day', 'week', 'month', or '12 months'. The exaSearch software can be utilized with any model whatever the model provider or creator. Exa returns web outcomes and extracted content material for agent workflows that want present info, area filters, date filters, and token-environment friendly excerpts. To use Exa Search, import gateway from ai and cross gateway.instruments.exaSearch() to the instruments parameter. When the mannequin needs current info, it calls the software and AI Gateway routes the request to Exa's Search API. Exa web search requests are charged at $7 per 1,000 requests. Each request includes up to 10 outcomes. Additional requested results past 10 are charged at $1 per 1,000 additional outcomes.

Search mode. Values: 'auto' (default), 'quick', or 'immediate'. Content category. Values: 'firm', 'folks', 'research paper', 'news', 'private site', or formal agreement 'monetary report'. Location: Two-letter ISO nation code, akin to 'US', for location-aware search. Domains: List of domains to restrict search outcomes to. Domains: List of domains to exclude from search outcomes. PublishedDate: Only return outcomes printed after this ISO 8601 date. PublishedDate: Only return outcomes published earlier than this ISO 8601 date. Controls extracted web page content and freshness. Return web page textual content. You possibly can set maxCharacters, includeHtmlTags, verbosity, includeSections, and excludeSections. Return concise excerpts. You can set question and maxCharacters. AgeHours: Maximum age of cached content material in hours. Timeout: Timeout for dwell crawling in milliseconds. Number of associated subpages to crawl. Target: Target page or pages for subpage crawling. Extract hyperlinks or image hyperlinks from pages with links and imageLinks. This preliminary AI Gateway integration helps Exa's customary Search modes and content extraction controls.

Deep synthesis modes and generated summaries usually are not uncovered but as a result of they've separate pricing. For extra details on search parameters and API options, see the Exa Search API documentation. The parallelSearch tool can be used with any mannequin whatever the mannequin supplier or creator. Parallel AI supplies LLM-optimized internet search that extracts relevant excerpts from net pages, making it excellent for analysis tasks and information retrieval. To make use of Parallel Search, import gateway from ai and cross gateway.tools.parallelSearch() to the instruments parameter. When the mannequin wants present information, it calls the device and AI Gateway routes the request to Parallel's search API. Parallel internet search requests are charged at $5 per 1,000 requests (contains as much as 10 outcomes per request). Additional results beyond 10 are charged at $1 per 1,000 extra results. Search mode preset. Values: 'one-shot' (comprehensive outcomes with longer excerpts, default) or 'agentic' (concise, token-environment friendly results for multi-step workflows).

Queries: Optional record of keyword search queries to complement the objective. Policy: Controls which domains and date ranges to include or exclude. Domains: List of domains to exclude from search results. Date: Only return results revealed after this date (format: YYYY-MM-DD). CharsPerResult: Maximum characters per result excerpt. CharsTotal: Maximum whole characters throughout all outcome excerpts. AgeSeconds: Maximum age of cached content in seconds for time-sensitive queries. For extra particulars on search parameters and API options, see the Parallel AI Search documentation. Use native net search instruments from Anthropic, OpenAI, Google, or xAI. These tools are optimized for his or her respective suppliers and may provide further features. Pricing for supplier-specific web search instruments will depend on the model you utilize. See the net Search worth column on the mannequin detail pages for precise pricing. For Anthropic models, you need to use the native internet search device provided by the @ai-sdk/anthropic bundle. 20250305() to the instruments parameter.