Tool use (also called function calling) is the capability for a large language model to invoke external functions or APIs to retrieve information or perform actions, rather than answering purely from its pretraining knowledge. The pattern is simple: the model is given a list of tools with their names, descriptions, and parameter schemas; when the model decides a tool would help, it emits a structured tool call; the application executes the call; the result is returned to the model, which incorporates it and continues. OpenAI shipped the first widely-used version in June 2023; Anthropic's tool use followed; by 2025 every major LLM API supports a nearly-identical interface, and the Model Context Protocol (MCP) has standardised the server-side contract.
Tool use transformed LLMs from pure text generators into agent components. **Retrieval** (query a vector store for relevant docs), **computation** (call a calculator for exact arithmetic), **live data** (fetch current weather, stock prices, flight availability), **action** (send an email, create a ticket, book a meeting), and **multi-step reasoning** (chain several tool calls) all become possible with the same primitive. The ReAct pattern formalised the reason-then-act loop that most agent frameworks use under the hood. LangChain, LlamaIndex, CrewAI, AutoGen, and vendor-native assistant APIs (OpenAI Assistants, Anthropic tool use, Gemini function calling) are all implementations of this basic idea.
For APAC mid-market teams building their first production AI workflows, tool use is usually the bridge between "chatbot that sometimes hallucinates" and "system that reliably gets work done". The moment you add a tool that returns authoritative data — your CRM, your product catalogue, your pricing engine, your knowledge base — the model stops needing to remember or guess. It just needs to decide when to ask.
The non-obvious operational note: **tool descriptions are part of the prompt and dominate model behaviour**. Vague descriptions produce vague tool use (the model hesitates, asks irrelevant follow-ups, or tries the wrong tool). Precise descriptions with clear distinguishing text for each tool — "Use this when the user asks about pricing for a specific SKU" rather than "get pricing" — dramatically improve first-call accuracy. Treat tool descriptions as product copy, not API documentation.
Where AIMenta applies this
Service lines where this concept becomes a deliverable for clients.
Beyond this term
Where this concept ships in practice.
Encyclopedia entries name the moving parts. The links below show where AIMenta turns these concepts into engagements — across service pillars, industry verticals, and Asian markets.
Other service pillars
By industry