Tag archive
CVE-2026-76066: IntentKit's executor validated tool calls against the full registry instead of the per-request binding, letting public callers invoke owner-only tools by name.
从零手写 Agent（四）：多 Tool 与联网检索 前面三篇，我基本把一个 Agent 最核心的骨架搭出来了： 第一篇：Tool...
从零手写 Agent（一）：Tool 把我从 Chat 拽进了 Loop 最近开始自己折腾 Agent。 一开始我也没急着上 LangGraph 之类的框架，而是在本地用...
Tools give a text model hands. MCP makes every tool speak the same language.
Dal prompt singolo ai grafi di lavoro, fino a sistemi che ottimizzano se stessi: una mappa pratica...
To let an LLM call your NestJS code, decorate a service method with @Tool() and its parameters with...
Qwen3-14B vs Llama-3.2-3B on real function-calling tasks. JSON validity, schema correctness, multi-turn handling. The smaller model won on cost and reliability; the bigger model won on accuracy. Here are the numbers.
The shift from generating the whole function to an agent that reads, searches, edits, and runs with tools: the ReAct loop and why the agent builds its own context instead of receiving it in the prompt.
Stop parsing prose with regex. Learn how to make an LLM return clean JSON on a schema you control, and how to let it call your functions safely with t
Spring AI released a fix on 21 August for a bug where a tool absent from the request could still be...
Small language model vs LLM: Bengaluru's Kinetic-4B beats Claude Haiku 4.5 at tool calling with 82.33% accuracy and 2.5x lower p95 latency.
ADS required capability names to contain a dot. OpenAI, Anthropic and Google all reject a dot in tool function names. No conformant name was usable as a tool. How we found it and what ADS-2 changed.