Tool use
Tool use is when a language model decides to call an external function (a tool) and uses the function's output to shape its response. Tool use is what turns a chat product into an agent that can read your email, send your invoice, or book your appointment.
Why it matters
A language model on its own can describe how to send an email but cannot actually send one. Tool use is the bridge between the model's reasoning and real-world action. The model sees a request ("send the invoice"), picks the right tool from a registered catalog ("send_invoice"), supplies the arguments ("customer X, amount Y"), and lets the runtime execute. The tool returns a result; the model continues reasoning with that result in context.
Without tool use, the chat product is a text generator. With tool use, the product is an agent that can read, write, look up, send, and book. The shape difference is dramatic: the surface goes from "describe what you would do" to "do it."
The engineering challenge is not the calling itself; it is everything around it. Tool authentication. Tool failure recovery. Tool sandboxing. Approval gating for consequential actions. Logging for an audit trail. The tool layer is where reliability either lives or dies; a flaky tool is worse than no tool because the user thinks something happened when nothing did.
How Ajeris handles this
Ajeris uses the Model Context Protocol as the contract between the agent and tools. Built-in tools cover the 21 personal-side OAuth integrations and the business-agent surface. Custom MCPs let users plug in any HTTPS endpoint. Approval gating fires before any consequential action (money movement, public posting). All tool calls write to a logged trail the user can review.
