Definition
Tool use is Tool use, also called function calling, is when an AI model asks the software around it to run an action, such as a search or a calculation, and then uses the result in its answer.
Tool use, also called function calling, lets an AI model do more than write text. The developer describes a set of tools to the model, such as "search the web," "read this file," or "look up an order." When the model decides one would help, it does not run the tool itself. It outputs a structured request naming the tool and its inputs. The software around the model runs the tool and hands the result back.
A useful picture is a chef who cannot leave the kitchen. The chef writes an order for the shopping assistant, the assistant fetches the ingredients, and the chef keeps cooking with what arrives.
A typical exchange goes like this:
Repeat that loop many times with a goal attached and you have an AI agent. The code that runs the loop and executes the calls is the agent harness. Standards such as the Model Context Protocol exist so a tool built once can be used by many different models and apps.
Tools are how models touch the real world. They fix things models are bad at on their own, like exact arithmetic, current facts, and anything that needs your data. Retrieving documents for an answer, as in retrieval-augmented generation, is one kind of tool use. Writing to a calendar, sending an email, or editing code are others.
That power is also the risk. A tool call can change something, spend money, or leak data. Good setups limit what each tool can do, ask for approval on risky actions, and run inside an agent sandbox. Text a tool returns can contain hidden instructions, which is the route for prompt injection.
When you judge a product that claims tool use, ask which tools it has, what each one is allowed to do, who approves actions, and whether you can see a log of every call. A model that picks the right tool and fills in the inputs correctly is doing real work. One that guesses wrongly can fail quietly, so the log matters more than the final answer.
Function calling is another name for tool use. The model outputs a structured request to run a named function with certain inputs, and the surrounding software runs it and returns the result.
No. The model only asks for a tool call. The application around it decides whether to run it, executes it, and sends the output back to the model.
Mistral released Voxtral 3 on Oct. 20: new transcription and speech models plus a voice agent API at $0.04 a minute when it reasons with Mistral Small 4. Skip the benchmark charts. The per-minute price and the choice of where audio is processed are the story.
Amazon is in talks to buy Decart, whose software ports AI models between chips, for about $7 billion, the Wall Street Journal reported Sunday. Read it as a bet on Amazon's Trainium chips, and on making your workload easier to move onto them.
Reflection AI introduced Beam on Monday, a 501-billion-parameter mixture-of-experts model with 23 billion active, and promised Apache 2.0 weights later in October. Until the weights ship, it is a waitlist and a claim about efficiency.
OpenAI says a limited group of US advertisers will start placing labeled visual ads alongside ChatGPT image generation later in October. It picked the most commercial moment in the product, and the new measurement tools say ChatGPT is becoming an ad platform.
DHS published notice on Oct. 20 of an order that bars DeepSeek from civilian agency purchasing and orders agencies to remove it, including third-party products built using DeepSeek. If you sell to government on a DeepSeek model, that clause is aimed at you.