When the agent searches
The agent decides when a search would improve the result. You don’t need to tell it to search, it happens automatically when needed.Looking up libraries or frameworks
When you mention a tool, library, or framework the agent isn’t familiar with, it searches for documentation and examples. Example: You say: “Add Stripe payment integration” The agent searches for current Stripe documentation and implementation guides to ensure it uses the right approach.Checking current best practices
When building features where approaches may have changed recently, the agent searches for up-to-date methods. Example: You say: “Set up authentication” The agent searches for current authentication patterns and security best practices to implement it correctly.Finding current information
When your app needs facts, data, or content that changes over time, the agent searches to get accurate information. Example: You say: “Add a section about the latest AI trends” The agent searches for current information instead of using outdated knowledge.How searches work
The agent formulates a search query based on what it needs, retrieves results from the web, and uses the information to inform what it builds.Quick searches
Most searches are quick and lightweight. The agent gets general information, overviews, or basic facts. This works well for:- General concepts
- Quick reference information
- Confirming basic details
Deep research
Sometimes the agent does deeper research when it needs comprehensive technical information. This happens when:- Looking up detailed API documentation
- Finding installation instructions for a package
- Understanding complex technical concepts
- Regular search didn’t provide enough detail
When searches aren’t needed
The agent doesn’t search for everything. For common tasks, standard features, or changes to existing code, the agent already knows what to do. It only searches when external information would improve the result. Simple requests like changing colors, updating text, or fixing obvious bugs don’t require searches. The agent works directly with your code.How this helps you
Web search means the agent can build with current information. Works with new toolsEven if a library or framework was released after the agent’s training, it can search for documentation and learn how to use it. Stays up to date
Best practices and approaches change. The agent searches to find current methods instead of using outdated patterns. Finds accurate information
When building content that includes facts or data, the agent searches to ensure accuracy rather than guessing or using old information. You get better results because the agent isn’t limited to what it knew at the time of training. It can look up what it needs, when it needs it.