Will an agent call a tool multiple times?

[NCCYUNSONG]

If is defined a web search tool can search one query a time. Then in the agent prompt there is " " - Use web_search_tool for all additional information if needed.\n"" If the agent has multiple questions, will he call this tool multiple times until all questions are answered?

[muhammadahmad1857]

✅ Yes, the agent will call the same tool multiple times if the user prompt logically requires it — even if there’s no error involved.

In the OpenAI Agents SDK (2025), the agent loop lets the LLM reason step-by-step. If the user asks something like:

“Get me the products in both electronics and clothing.”

…and your get_products tool takes only one category at a time, the agent will internally:

1. Call get_products(category="electronics")

2. Then call get_products(category="clothing")

3. Finally, combine and return the results

You don’t need to manually split the task — the LLM can handle this multi-call pattern by itself, as long as your instructions make it clear:

"Use the tool multiple times if needed to gather all required information.”

🔁 This works whether the tool is web_search_tool, get_products, or any other function-based tool.

📌 Just make sure:

Your tool is designed to accept one input at a time (e.g., one category)

The agent is allowed to make multiple tool calls per run

Hope that helps!