Guides and price comparisons for giving AI agents the web: search, extraction and browsers.
Search APIs used to return ten links. The best ones now read the pages, and four clear 95% on SimpleQA. browserfoundry deep search scores 95.30% with no model in the loop, at $3 per 1,000 requests: the lowest price of the four. Every score, price and record, sourced.
→Engineering · October 7, 2026Inside the search tier: admission control, a freshness-classed cache with request coalescing, deadline propagation, latency-aware routing and hedged requests that take the p99 from 1.48 s to 587 ms. With the numbers, the code and the incident that shaped it.
→Search APIs · October 5, 2026Google's Custom Search JSON API is closed to new customers and stops on January 1, 2027. What changes, what Google recommends, and the alternatives compared on price, limits and migration effort.
→Search APIs · October 5, 2026DuckDuckGo has no official web search API. Why the duckduckgo-search (ddgs) library keeps throwing "202 Ratelimit", how to make it hold up, and the free-tier APIs to move an agent onto.
→Extraction · October 5, 2026Turning a web page into clean markdown for an LLM: Python libraries (markdownify, html2text, trafilatura, MarkItDown), Jina Reader, Firecrawl and browserfoundry's extraction and rendering endpoints, compared on JavaScript, boilerplate, limits and cost.
→Search APIs · October 5, 2026Built-in web search for LLMs, priced side by side: OpenAI's web search tool, Claude's web search tool, Gemini grounding with Google Search and Perplexity Sonar, with the token costs most comparisons leave out and a worked monthly bill.
→Browsers · October 5, 2026Browserbase's Free, Developer, Startup and Scale plans decoded: browser hours, concurrency, proxies and overage, with monthly bills worked out and compared against Steel, Cloudflare Browser Run and others.
→