Interview OS
Back to Context Engineering

Library

Tool use & function calling

Let the model act.

Agents

Overview

Tool use lets an LLM call external functions — search, code, APIs — extending it beyond text generation. This topic covers defining tool schemas, parsing calls, and feeding results back into the reasoning loop.

How it works

Agents
DefineDecideExecuteOutputTool schemaspecModel picksJSON argsRun toolobservationAnswer
ClientServiceEdgeData

Step by step, with examples

  1. 1

    Tool schema

    • Name, args, and description.
  2. 2

    Model picks

    • The LLM emits a tool call.
  3. 3

    Run tool

    • Call the function; return the result.
  4. 4

    Answer

    • Loop until the task is done.
    • Example: function calling

Overview

Expose typed tools the model can call; design clear names, descriptions, and schemas so the model picks the right action.

Common pitfalls

  • Too many overlapping tools
  • Vague tool descriptions
  • No error feedback loop

Where this content comes from

For full transparency, this content is curated and verified from these sources:

Frontier-lab prompting & agent guidesRetrieval-augmented generation literatureOppZen-authored context-engineering playbooks