OpenRouter Launches Agentic Web Tools for Model Agnostic Search and Fetch

OpenRouterOpenRouter

· Updated

OpenRouter released server-side web search and fetch tools that allow any tool-calling model to autonomously browse the live web. By standardizing the tool schema, developers can swap between providers like OpenAI and Anthropic without rewriting their search implementation or domain filters.

OpenRouter, a unified API for accessing hundreds of language models, launched openrouter:web_search and openrouter:web_fetch as server-side tools. These tools enable agentic AI to decide when to search, what queries to use, and how many times to retrieve data within a single request.
Search engines
Exa, Parallel, p0, and others
Fetch engines
OpenRouter, Exa, Native, and others
Search pricing (Exa/Parallel)
$0.005 per request
Fetch pricing (Exa)
$0.001 per fetch
Fetch pricing (OpenRouter)
Free
Availability
API and OpenRouter chatroom

This update solves the fragmentation of model-specific search schemas. Previously, switching from GPT to Claude required rewriting how tools were defined and results parsed. Now, a single tool definition works across all providers, mirroring the platform's recent push into long-running agent primitives and model-agnostic routing.

You can now implement granular controls like domain filtering and token caps to manage costs and context window usage. The tools are available via the API and the OpenRouter chatroom. Developers using the legacy web plugin should migrate to the new tools parameter to enable multi-step loops similar to Perplexity's agent infrastructure.

View the full update on openrouter.ai
OpenRouter
OpenRouter
@OpenRouter
X

Any tool-calling model on OpenRouter can call web search and web fetch agentically. The model decides when to search, what to search for, and how many times. We've added @p0 as a new web search provider. Learn more: https://t.co/dDKZ24VcqN https://t.co/xlFK7FN9cf

2retweets46likes
View on X

Still wondering? A few quick answers below.

OpenRouter agentic web tools are server-side capabilities that allow any tool-calling model to autonomously search the web and fetch page content. Unlike legacy plugins that run a single search before generation, these tools let the model decide when to search, what queries to use, and how many times to retrieve data during a request.

These tools provide a standardized schema that works across all supported model providers, such as OpenAI, Anthropic, and Google. This consistency allows developers to swap models without rewriting their search implementation or parsing logic. Users can also choose specific search engines like Exa or Parallel to ensure identical search behavior regardless of the underlying model.

Pricing varies by the engine selected. Native provider search uses the model provider's own pricing. Using Exa or Parallel for search costs $0.005 per request for up to ten results. For fetching content, the OpenRouter engine is free, while the Exa engine costs $0.001 per fetch to provide cleaned markdown output from a URL.

Yes, both the web search and web fetch tools support domain filtering through allowed_domains and blocked_domains parameters. This feature is critical for developers building agents that must only interact with trusted sources or specific documentation. These filters are strictly enforced when using non-native engines like Exa, Parallel, or OpenRouter's own fetcher.

To migrate, you must replace the plugins field in your API request body with the tools field. Specifically, change the plugin ID web to the tool type openrouter:web_search. Note that this transition requires using a model that supports native tool-calling; models without tool-calling capabilities must continue using the legacy plugin architecture.

Every HeadsUpAI update is written based on its original source and reviewed before it's published. Read our editorial standards →

See all AI news & updates from OpenRouter →

Share this update