Can OpenAI's web search tool run without live internet access?

Definition: Yes. In the Responses API, set external_web_access to false on the web_search tool. OpenAI documents this as offline/cache-only mode, using cached or indexed results without fetching new external content.

Key facts

• OpenAI's web search documentation says the Responses API can control whether the tool fetches live content or uses cached or indexed results.
• The Responses API reference defines external_web_access as an optional boolean that defaults to true when omitted.
• The same reference says false runs web search in offline/cache-only mode and prevents fetching new external content.
• OpenAI documents a separate Agents API setting, mode: "cached", for searching saved web content without accessing the live internet.
• The legacy web_search_preview tool ignores external_web_access and behaves as if live access is enabled.
• In the Responses API, tool_choice: "auto" leaves the decision to search with the model.

What does external_web_access change?

The setting controls live access for the hosted Responses API web_search tool. OpenAI's API reference describes external_web_access as the switch that allows live internet access, with true as the default. Setting it to false selects offline/cache-only mode and prevents the tool from fetching new external content.

The documented request shape is:

curl "https://api.openai.com/v1/responses" \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer $OPENAI_API_KEY" \
  -d '{
    "model": "gpt-6-astra",
    "tools": [
      { "type": "web_search", "external_web_access": false }
    ],
    "tool_choice": "auto",
    "input": "Find when the Eiffel Tower opened to the public and cite the source."
  }'

OpenAI's guide describes this as choosing between live content and cached or indexed results. It does not state how fresh cached content is, how long it remains available, or what happens when a query has no usable cached result.

Is cached mode the same in every OpenAI API?

OpenAI uses different controls across API surfaces. The Responses API uses the external_web_access boolean. The Agents API documentation uses a mode field with live, cached, and disabled values:

"tools": [{ "type": "web_search", "mode": "cached" }]

In the Agents API, live is the default when mode is omitted, cached searches saved web content without accessing the live internet, and disabled turns off built-in web search. These names belong to the Agents API interface; a Responses API request uses external_web_access.

Does the setting force a search to run?

No. The Responses API guide says that, like any other tool, the model can choose whether to search based on the input prompt. With tool_choice: "auto", search is optional. When an application must run web search, OpenAI documents tool_choice: "required" or a specific web search tool choice.

The legacy web_search_preview path is a separate limitation. OpenAI says it does not support newer controls including external_web_access, and the guide states that preview variants ignore the parameter and behave as if external_web_access is true.

What about Chat Completions search models?

Chat Completions uses specialized search models. OpenAI documents that these models always retrieve information from the web before responding, while the Responses API treats web search as an optional tool. The Chat Completions search path does not support Responses API live-access control, so external_web_access is not a control for that integration.

Related terms

• ChatGPT search
• Does ChatGPT always search the web before answering?
• Answer attribution

FAQ

Can you use cached results without adding web search?

No. The cache-only setting belongs to the web_search tool configuration. If the tool is omitted, built-in web search is not enabled for the request.

Does external_web_access: false make every result current?

No. It prevents new external fetching and selects cached or indexed results. OpenAI's documentation does not specify a freshness period for those results.

Does web_search_preview support cache-only mode?

No. OpenAI documents that preview variants ignore external_web_access and behave as if it were true.

Sources

1. OpenAI, "Web search", accessed 17 October 2026
2. OpenAI, "Create a model response", accessed 17 October 2026
3. OpenAI, "Web search", accessed 17 October 2026

---

Check how a page reads to answer engines. Open the GEO score analyzer