Built-in tools
Built-in tools extend what the AI can do beyond generating text. When a tool is enabled, the model can call it automatically during a response whenever it judges the tool helpful — no special commands from you are needed. Results appear in the chat as a collapsible tool-use trace so you can see exactly what happened.
For custom tools that connect to your own HTTP APIs, see Dynamic tools.
Available built-in tools
Web search
Queries a search engine and returns ranked results with titles, URLs, and snippets. Use this tool to keep answers grounded in current events, recent documentation, or any information that may have changed since the model's training cutoff.
Code interpreter
Executes Python code in a secure, isolated sandbox and returns the output — including printed text, data tables, and generated charts. Ideal for data analysis, numerical computations, file transformations, and any task where running code produces a more reliable answer than reasoning alone.
Calculator
Evaluates mathematical expressions in a safe sandboxed environment. Enabling the calculator prevents the model from making arithmetic errors in multi-step calculations, unit conversions, and financial computations.
Document reader
Parses and extracts text from files attached to the conversation — PDF, DOCX, CSV, and plain-text formats are all supported. The model can then answer questions, summarise content, or analyse data drawn directly from your documents. See File uploads for how to attach files to a chat.
Image handling
Lets the model inspect and reason about images you attach to a conversation. Works with screenshots, diagrams, photos, and charts — the model can describe, compare, extract text, or answer questions about the visual content.
Date and time
Returns the current date, time, and UTC offset so the model always knows exactly when "now" is. Particularly useful when you want the model to reason about deadlines, schedules, or time-sensitive information without relying on its training-data timestamps.
Enabling and toggling tools
In a chat
Open any chat and click the wrench icon in the toolbar above the message input. A panel lists every available tool with a toggle switch next to each one. Flip a toggle to enable or disable that tool for the current conversation. Changes take effect on your next message.
On an agent
When creating or editing an agent in the Agent builder, the Enabled tools section lets you select which tools the agent can always use, regardless of per-conversation settings. See the Agents guide for full details.
How the model uses tools
You do not need to ask the model to use a tool explicitly. When a tool is enabled and the model determines it would improve the answer, it calls the tool on its own, receives the result, and incorporates it into the final response. A single reply can involve multiple tool calls in sequence — for example, searching the web and then reading a linked document.
Every tool call is shown as a collapsible trace in the chat. Expand it to see exactly what the model requested and what result came back.
Enable only what you need
Enabling fewer tools keeps the model focused. If you are working on a task that does not require web access, turning off web search helps the model rely on the conversation context rather than retrieving potentially irrelevant external content.Custom tools
Need to connect the model to your own API or a third-party service not covered by the built-in tools? The Dynamic tools builder lets you define custom HTTP tools directly in the app — no code required.