Agent++ is a conceptual framework designed to enable safe, efficient, and standardized agentic automation by leveraging large language models (LLMs). It consists of two primary components:
-
Language Specification: A conceptual programming/scripting language tailored for LLMs, drawing inspiration from Python's flexibility and Bash's task-oriented syntax. The language is optimized for LLMs to learn, generate, and execute commands for a wide range of automation tasks.
-
Interpreter/Runtime Environment: A safety-centric execution layer designed to sandbox and execute commands generated in the Agent++ language. This runtime prioritizes security, allowing configurable safety settings and constraints to mitigate risks in autonomous task execution.
Agent++ is designed with three key goals:
- Adoptability: The language can be seamlessly integrated into LLMs through training and tooling.
- Automation-First Design: It supports agentic tasks and workflows effectively.
- Safety and Security: The runtime environment ensures safe, accountable execution of automated tasks.
As LLMs become increasingly capable, there is a growing demand for integrating them into autonomous workflows. However, existing solutions lack standardization, scalability, and robust safety mechanisms. Agent++ aims to fill this gap by providing a unified framework for agentic automation, combining a language optimized for LLMs with a secure execution platform.
Agent++ provides the foundation for reliable, safe, and efficient agentic automation, bridging the gap between powerful AI capabilities and responsible deployment. By focusing on both language design and runtime safety, it seeks to standardize and advance the use of LLMs in agentic roles.