You.com

Most AI scientist tools start from a blank page. Chembricks starts from the state of the art, and validates everything afterwards. Their AI scientist takes on prompts like this one: design a liquid that stores hydrogen fuel and releases it on demand, holding at least 5.5% hydrogen by weight (the US Department of Energy target), with a release energy between 9.6 and 16.7 kcal per mole of H₂. Before it proposes a single molecule, it uses You.com web search to ground itself in the current state of the art research. Then it runs its own work: running Chembricks quantum physics based AI models to predict internal energies, comparing options like N-ethylcarbazole against 1,3,5-triazine on thermal stability. Web search is one tool in that agent, but it plays a specific role: It grounds the AI scientist to what chemists have already published, tried, and validated,  so it starts from real ground instead of guessing. Chembricks is building the quantum physics validation layer for AI driven science: an AI scientist trained with expert verified quantum chemistry data that predicts molecular properties in milliseconds. We're proud to partner with web intelligence that keeps that science grounded in the SOTA scientific literature. See it work: chembricks.ai

Claude ships with basic web search. It holds up until your agent needs dense, current context mid-loop, and then retrieval becomes the slowest step in an otherwise fast workflow. You.com now runs in both places you already work: a connector in Claude Desktop, and a plugin in Claude Code. You get real-time search, full page extraction, and cited multi-tier research, all on the same production index that powers our API. Setup is one step either way. Search the connector directory for You.com and click Connect, or run /plugin install youdotcom-agent-skills@claude-plugins-official in Claude Code. Install it and stop treating web search like an afterthought in your agent's tool belt.

Join us, CrewAI, Daytona, and One Friday in NYC for a hackathon! We're spending a full day focused on building agents—but not just any agents. Self-repairing and self-learning agents. You get the You.com Search, Research, and Deep Research APIs, Daytona sandboxes so your agent can run its own code without taking down your laptop, CrewAI for orchestration, and One's four-command CLI for auth across 729 integrations. 1st place walks away with credits—$1k You.com, $3k Daytona, and $1k CrewAI—plus a year of One. Every builder leaves with credits regardless of where they place. Request to join below. 👇

New: Web Search with Highlights. 95.17% on SimpleQA. Same $5 CPM. Independently verified by Braintrust: 35% cheaper than Claude's built-in search, 11% cheaper than OpenAI's. Fast, really accurate, or cheap? We picked all 3. At 695ms p50, Highlights is up to 3x as fast as the competition with top tier accuracy. The full methodology and public harness are linked below, run it against your own workload. Which of the three do you trade away most often when you pick a search API? One field, no migration. Read the full breakdown, link in comments.

It’s common to see "cited answer" demos fall apart the first time someone asks a real diligence question. In collaboration with Pydantic AI, we show a version that doesn't fail: an M&A deal desk built from three specialist agents, each grounded in a different research surface. → One scans the market and press with domain boosting instead of a blunt allowlist.  → One pulls financials from a curated finance index, cited to filings and transcripts.  → One runs an adversarial risk pass pinned to sec.gov, justice.gov, and courtlistener, so open-web allegations can't pass as findings. A lead agent routes each to a fast or deep model, caps it with a timeout and a call budget, and synthesizes a typed memo where every red flag traces to a source. The whole run is one trace, so the reasoning holds up when a reviewer asks "how do you know that?" This is the new standard for a diligence agent: not just a good answer, a defensible one. You.com is now available as capabilities in the Pydantic AI harness. Full architecture and code in the comments.

Braintrust ran an independent eval and found that search configuration moves agent accuracy about 40x more than model choice does. 4 models, 1,329 questions. With the same retrieval layer, the accuracy gap between the cheapest and most expensive model narrowed from 47.9 points to 5.6. The cost story points the same direction. Same model, same questions: Claude Sonnet 5 with its built-in search costs $0.1148 per query. With You.com, $0.0747. 35% cheaper, comparable accuracy. Which are you optimizing first today, your model or your retrieval layer?

AI should answer questions, not blow smoke. Grateful to the KRAZAM team and Benjamin Burke for creating this video, highlighting that very problem. Cut to the chase. Get the right answers—quickly—with You.com web search APIs.

Our COO, Alex Triplett, sat down with Gemma Allen at theCUBE + NYSE Wired this week to talk about where enterprise search is headed. We’re a search lab built for the agentic web. Alex talked through a pattern we're seeing across enterprise conversations: companies are unbundling the stack and pulling web search out of their LLM instead of assuming the model handles it. Why? Zero data retention on search, real multi-model flexibility instead of lock-in to one provider, and the ability to optimize search and inference separately instead of paying for both bundled into one cost. You.com sits below the agent layer, powering search for agent-native businesses, frontier labs, and developers building on the web. Thanks to Gemma, John Furrier, Brian J. Baumann, and the SiliconANGLE & theCUBE and NYSE Wired team for having us. Full conversation linked below. 👇 #NYSEWired

No web search API returns more accurate results faster than You.com Web Search with Highlights. Artificial Analysis benchmarked 14 search APIs across BrowseComp, DeepSearchQA, and their Omniscience set. We score 74 on their composite Search Index and get there in 18 seconds per task. No other provider beats this accuracy without doubling your user’s wait. Link in comments. 👇

Products That Count picked 30 people out of 5,000 nominees for this year's Product Builder Awards. Saurabh Sharma, our CPO at You.com, is one of them. The award exists because the PM job has quietly changed. The strongest builders right now aren't the ones furthest from the work reviewing decks and roadmaps—they're the ones closest to it, in the codebase, in the prompts, in the actual product. Ownership over oversight. Congrats, Saurabh! 🎉 Good company to be named alongside too—product leaders from Stripe, Google, PayPal, and Amazon also made this year's list. 👇