2 posts tagged with "modes"

An AI agent that can read files, run shell commands, and hit external APIs is useful. An AI agent that does all of that when you only asked it to answer a question is a problem.

Modes are how Toolpack SDK draws that line. A mode defines three things: which tools the AI can access, what system prompt shapes its behavior, and whether the workflow engine (planning, step execution, retries) is active. You set the mode once at init or switch it at runtime, and the SDK enforces the constraints on every request.

By default, the Toolpack SDK provides a unified AI agent runtime with full access to an extensive toolset. But giving an AI full runtime access is not always the correct architectural choice. Sometimes you need a chat-only assistant, a read-only DevOps reviewer, or a highly specialized agent confined to a strict workflow.

This is exactly what Modes solve. A "Mode" shapes the AI's persona, its capabilities, and its operational boundaries. In this guide, we will explore what modes do, how the built-in modes work, and how to create powerful custom ones from scratch.