Skip to main content

Documentation Index

Fetch the complete documentation index at: https://ai.aidalinfo.fr/llms.txt

Use this file to discover all available pages before exploring further.

Tools let your agents trigger application code (search, internal APIs, business logic…). AI Kit registers them on the agent and injects them into every generate or stream call.

​
Register default tools

import { Agent, google } from "@ai_kit/core";

const assistant = new Agent({
  name: "life-assistant",
  instructions: "Lifestyle assistant",
  model: google("gemini-2.5-flash"),
  tools: {
    google_search: google.tools.googleSearch({}),
  },
  loopTools: true,
  maxStepTools: 15,
});

await assistant.generate({ prompt: "What’s the weather in Paris?" });
  • The tools option exposes tools for every call.
  • loopTools and maxStepTools defined on the agent act as defaults.
  • You can override any of them per call (agent.generate({ tools, loopTools, maxStepTools })).

​
Automatic tool loop

When loopTools is enabled and at least one tool is available, AI Kit orchestrates the loop: after each toolCall, the corresponding tool executes and the conversation resumes until a final answer is produced. The result includes a loopTool flag for quick DX inspection. 
const result = await assistant.generate({
  prompt: "Find the workflow documentation for the tool loop feature.",
  loopTools: true,
});

if (result.loopTool) {
  console.log("Tools were executed during the conversation.");
}

​
Control knobs

  • loopTools?: boolean – enable or disable the loop (disabled by default).
  • maxStepTools?: number – cap the total amount of tool executions (default: 20).
  • Override both on individual calls when you need a specific behaviour.
await assistant.generate({
  prompt: "Reply without executing any tool.",
  loopTools: false,
});

​
Recommendations

  • Prefix tool IDs (search.customer, support.lookup) to simplify monitoring.
  • Lower maxStepTools when a tool can trigger expensive side effects.
  • Log toolCalls in your observability stack to audit model decisions.