Skip to content

OpenAI Built-in tools

OpenAI Built-in tools gives you access to the tools OpenAI has built in: searching the internet, generating images and analysing a data file. You write a request, the OpenAI model uses the chosen tool and returns the result: a written answer with its sources, an image, or the file analysis.

You can use it as a step in a flow, with the request in Input, or as a tool for an agent: connect its Tool output to the agent’s Tools input and the agent uses it when it needs to search, create an image or analyse the file.

When to use it

  • An agent needs current information from the internet and you want to limit it to certain websites.
  • You want search results to take a specific country or city into account.
  • A marketing flow has to generate images from a description.
  • You need AI to answer questions about a spreadsheet or a PDF, such as totals or trends.

Configuration

You need an OpenAI API credential with your OpenAI account key. See Credentials.

FieldWhat it does
Built-in toolThe tools to use: web_search (search the internet), web_search_preview (legacy) (OpenAI’s previous search, without options), image_generation (generate an image) and code_interpreter (analyse a file). Default: web_search. Ticking an option shows its own fields.
InputThe request, for example Latest news on digitalisation grants for small businesses. It is hidden when you connect the Tool output, because the agent writes the request.
OpenAI API KeyYour OpenAI account key.
ModelThe OpenAI model that answers. Default: gpt-5-mini. The list shows the models in your account.

Shown when you tick web_search.

FieldWhat it does
Search Context SizeHow much information from the pages found the model uses: low, medium or high. More information gives fuller answers and uses more. Default: medium.
User Location (Country)Country to tailor the results to, as a two-letter code, for example US or ES.
User Location (City)City to tailor the results to, for example Madrid.
User Location (Timezone)Time zone, for example Europe/Madrid.
Allowed DomainsWebsites the search is limited to, separated by commas, for example gov.uk,hmrc.gov.uk. When empty, all are searched.

Image generation

Shown when you tick image_generation.

FieldWhat it does
Image QualityImage quality: auto, low, medium or high. Default: auto.
Image SizeSize: auto, 1024x1024 (square), 1024x1536 (portrait) or 1536x1024 (landscape). Default: auto.
Output FormatFile format: png, webp or jpeg. Default: png.
BackgroundBackground: auto, transparent or opaque. Default: auto.

Advanced options

FieldWhat it does
File for Code InterpreterOnly with code_interpreter. The file to analyse: CSV, Excel, JSON, text or PDF.
Assistant InstructionsOnly with code_interpreter. Instructions for the analysis. They come filled in with instructions to reply with the final answer only.
Tool NameThe name the agent sees for the tool. Default: openai_built_in_tool.

Output

  • Data: depending on the tool used:
    • Web search: the written answer, with its citations and the sources consulted.
    • Image generation: the name and link of the image created, which is saved in the conversation files.
    • File analysis: the answer about the file’s content.
  • Tool: the tools ready for an agent, one for each ticked option. Connect it to the agent’s Tools input.

Example

A communications agency wants an assistant that prepares posts with an image.

  1. Add OpenAI Built-in tools and tick only image_generation.
  2. In Image Size, choose 1024x1024, and in Output Format, png.
  3. Connect the Tool output to the agent’s Tools input.
  4. When someone asks for “An image of a cup of coffee on a wooden table for Monday’s post”, the agent generates the image and returns its link.

Good practices

  • When you connect it to an agent, tick only one tool per component. If you need several, add one component for each and give them different names in Tool Name.
  • code_interpreter can’t be combined with other tools in the same component when you use it as a flow step.
  • Images and searches use credit from your OpenAI account. Check your usage there.