Tool Use
The ability of an LLM to request that the surrounding application run a function, API call or query, then use the returned result to continue its answer.
Definition
An LLM on its own can only produce text. Tool use gives it a way to reach beyond that. The application describes a set of tools to the model: a name, a plain-English description and a schema for the arguments. When the model decides a tool would help, it emits a structured request instead of prose. The application runs the tool, returns the result as a new message, and the model carries on, now with real data in hand.
Tools can be anything with a defined input and output: a database query, a web search, a calculator, a ticketing API, a code sandbox. Providers expose the mechanism under slightly different names; "tool use" (Anthropic) and "function calling" (OpenAI and others) describe the same pattern. Chained tool use is the engine inside every AI agent.
Example: a user asks "is order 88-1204 delayed?". The model calls get_order(id="88-1204"), receives a JSON status of "held at customs", and replies with that fact and the expected date, instead of guessing.
What it means in practice
- It turns an assistant that talks about your systems into one that reads and updates them.
- Answers become grounded in live data, which cuts hallucination for factual queries.
- Each tool is an attack surface for prompt injection; permissions belong in the tool rather than the prompt.
- Good tool descriptions are product work: vague tools get called wrongly or not at all.
How Teksmobile uses it
Designing the right tool set, with narrow scopes and validated inputs, is a central part of our AI agents and automation engagements. We package tools as MCP servers so they can be reused across assistants, and we connect them to existing systems through our AI integration service.
Related terms
Part of the Teksmobile AI Glossary. For longer reads, see our guides.
Frequently asked questions
Does the model execute the tool itself?
No. The model only produces a structured request. Your application decides whether to run it, runs it, and returns the result. That separation is where you enforce permissions.
How many tools can a model handle?
Dozens work in practice, but accuracy falls as the list grows and descriptions blur. Group tools by task, write precise descriptions and expose only what the current workflow needs.
Is tool use the same as plugins?
Plugins were a product feature of specific chat products. Tool use is the underlying API mechanism, and MCP is the open standard for packaging tools so any application can use them.