Web search

Give an agent a web_search tool and it can fetch fresh results from the open web. No extra API keys — Shmastra routes the search through the same LLM provider the agent already uses.

Example prompts

Give the research agent a web-search tool.

Add web search to the sales-advisor — it should look up companies before drafting emails.

Equip this agent with web search.

What happens

The widget updates the agent's tool list with createWebSearchTool(<agentId>):

import { createWebSearchTool } from '../shmastra'

export const myAgent = new Agent({
  id: 'my-agent',
  tools: createAgentTools({
    web_search: createWebSearchTool('my-agent'),
  }),
})

A dry-run build confirms it compiles, and the agent reloads. Next time you chat, it can search.

Using it in practice

• "Find the latest funding round for Anthropic."
• "Compare current pricing for the top three cloud GPU providers."
• "Check if there's a new release of Mastra and summarise the changelog."

Tips

• Say when to use it. Add to the agent's instructions: "Prefer web search for anything that might have changed in the last six months." — otherwise the agent may rely on training-data memory.
• Combine with the browser. Web search gives a result list; headless browser fetches the full page.
• Cache expensive queries. For repeat research tasks, consider a workflow with a web-search step whose output you store.