We built it internally at first, when working on AI tools. Got tired of messy web retrieval in most LLM workflows - and built our own stack for it.
Right now, web search for AI means chaining together a search API, scraping, captcha handling, HTML cleanup, content extraction, prompt formatting. And the models still end up processing cookie banners/menus/nav text/junk. Cue inflated token usage, bad latency, and inconsistent results.
Our Search Router is built specifically for AI agents and RAG systems. We made sure it is fast, the output is easy to work with, and raw HTML is avoided wherever possible. Recently we added a “Retrieved Context” endpoint that scrapes pages and extracts relevant content into structured JSON, saving time and tokens on cleanup.
We also have native MCP support, so you can directly plug Search Router config into Claude Desktop and let it search through the tool without burning Anthropic limits.
For launch, we have a free tier (no card required) with effectively unlimited usage - good for testing.
Thanks in advance for feedback from people building agents, RAG pipelines, or other LLM products. What's your solution for AI web search now and how does Search Router compare?