Tool Use
Core Facts
Function Calling (Tool Use) is a standardized mechanism for large models to invoke external tools, where developers pre-declare tool names, descriptions, and parameter schemas, and the model autonomously determines when to make a call.
The underlying mechanism of Tool Use is that developers pre-define the name, description, and parameter structure (Schema) of tools, and the model determines during inference when and which tool to call and outputs structured invocation instructions.
The tool-calling (Function Calling) mechanism allows the model to invoke external tools during reasoning, such as file read/write operations, terminal commands, and search APIs.
Timeline (last 90 days)
Modern coding agents output structured tool-call requests (such as read(path) and shell(cmd)), which are executed by an external runtime that returns the results; the model itself does not have the ability to directly read or write files.
When evaluating Agentic Coding, tool call success rate, call count and error recovery ability often reflect actual engineering value better than single-turn completion scores.
The minimum threshold for getting started with AI Agent development includes: Python fundamentals, RAG (Retrieval-Augmented Generation), Tool Use (tool calling), and completing a full project using LangChain.
The difference between an Agent Loop and a single API call in tool calling is that a single call has no tool-calling capability and can only output text, while an Agent Loop can call tools such as the file system, databases, and external APIs.
The advanced section moves into RAG knowledge bases, Agent architecture, multi-Agent collaboration, and tool calling, with development using mainstream frameworks.
In classic paradigms such as ReAct, tool-call decisions are made by the same large model, and when the task chain is long the loop can run dozens to hundreds of rounds, with latency and cost accumulating linearly.
Before ChatGPT came out, the roadmap already listed research directions such as Coding Agents, Tool Use, long context, RL, automated feedback, and Agent safety.
Current mainstream LLM agents do not execute code or SQL directly; instead, the model outputs structured text (usually JSON) declaring the functions and parameters to be called, which the host program then parses and executes.
When a model enters agentic mode and makes autonomous decisions across multiple steps through tool calls, each step's output becomes the input for the next step, and errors or deviations can be amplified exponentially within the loop.
Apple's Tool protocol shares a common design philosophy with OpenAI's Function Calling and Anthropic's Tool Use, all of which decouple the LLM's natural language understanding capability from deterministic program logic.
30 more timeline events
All Facts (20)
Function Calling (Tool Use) is a standardized mechanism for large models to invoke external tools, where developers pre-declare tool names, descriptions, and parameter schemas, and the model autonomously determines when to make a call.
90%VerifiedThe underlying mechanism of Tool Use is that developers pre-define the name, description, and parameter structure (Schema) of tools, and the model determines during inference when and which tool to call and outputs structured invocation instructions.
90%VerifiedThe tool-calling (Function Calling) mechanism allows the model to invoke external tools during reasoning, such as file read/write operations, terminal commands, and search APIs.
90%VerifiedOpenAI's Function Calling and Anthropic's Tool Use API are both standardized implementations of the ReAct pattern.
90%VerifiedAI Agent通过工具调用(Tool Use)机制与外部系统交互,包括执行代码、调用API、读写数据库等
90%VerifiedAnthropic and OpenAI successively released models supporting tool use and multi-step reasoning around 2024.
90%VerifiedThe operation of AI Agents relies on large language model APIs, vector databases, tool-calling interfaces, third-party SDKs, and various rapidly iterating frameworks, resulting in an extremely long execution chain
80%VerifiedTool-Augmented AI is an important development direction for current large model applications, with mechanisms such as OpenAI's Function Calling and Anthropic's Tool Use underpinning this paradigm at the foundational level.
80%VerifiedFunction Calling机制中,开发者将工具的名称、功能描述、参数Schema(通常为JSON Schema格式)注入到模型的请求中,模型输出结构化的JSON而非自然语言来调用工具
80%VerifiedAI Agent与传统聊天AI的根本区别在于具备感知-决策-行动的闭环能力,采用ReAct(Reasoning + Acting)框架
80%VerifiedAnthropic introduced Tool Use in Claude, and Google provides a capability similar to Function Calling in Gemini.
80%VerifiedAnthropic introduced Tool Use as its Function Calling solution.
80%VerifiedFunction calling (tool calling) refers to the ability of a large language model to proactively request the invocation of external functions or APIs during the generation of a response.
80%VerifiedTool Use implementations from different model providers vary in parameter formats, parallel call support, streaming tool calls, and other aspects, and the AI SDK's unified abstraction layer is used to shield these differences.
75%VerifiedThe basic principle of tool calling is that the model identifies scenarios requiring external capabilities, outputs a structured function call request (usually in JSON format), and after the external system executes it, the result is returned to the model to continue reasoning.
75%VerifiedResearch indicates that models may produce unfaithful chains of thought, where the surface-level reasoning steps are inconsistent with the actual basis for decision-making.
75%VerifiedCoding Agent与普通AI聊天机器人的本质区别在于工具调用(Tool Use)能力,包括执行终端命令、读写文件系统、调用API等
75%VerifiedTool calling allows models to dynamically invoke external functions during inference and incorporate execution results into the context of subsequent reasoning, forming a closed loop of perception, decision-making, and execution.
70%VerifiedAnthropic's Tool Use technology provides the underlying support for the Agent loop, enabling the model to call external tools during the reasoning process and continue reasoning based on the returned results.
70%VerifiedAI Agents interact with external tools through standardized interfaces (such as OpenAI's Function Calling and Anthropic's Tool Use protocol), and CLI command-line tools are naturally suited to this invocation pattern.
70%