I built Chloe: an open-source TypeScript framework for AI agents you control

I developed Chloe, an open-source TypeScript framework designed for creating AI agents that businesses, particularly small ones, can control. The framework's concept revolves around employing code for predictable tasks while utilizing AI for necessary operations. For instance, an agent managing an invoice would leverage AI to interpret the invoice, apply coding to verify financial details and adhere to business rules, and finally seek human approval before processing the payment.

The idea is to utilize regular code for tasks that require predictability and avoid consuming AI tokens for unnecessary work. Chloe allows developers to integrate both code-driven and AI-driven steps using TypeScript.

Beyond its functionality, Chloe aims to provide developers with greater control over the agents they create. By using Chloe, developers maintain ownership of the code, tools, skills, and tests that validate the agent's performance. This framework supports running agents on personal machines or servers while determining which external services the agents interact with.

However, it's crucial to note that running locally doesn't guarantee privacy, especially when using cloud-based AI models, as the data sent to these models still leaves the machine.

Throughout my career, I've collaborated with businesses as a forward-deployed engineer, tailoring tools to meet their specific requirements. This experience heavily influenced the development of Chloe. My primary goal was to create an agent framework that developers could fully comprehend, thoroughly test, and modify before deploying it to a client or their own business.

Chloe is currently available for free and open source, reflecting my intention to make AI agent creation more accessible. Should you be involved in developing agents for clients or your own business, I would appreciate your insights on which aspects prove most challenging to ensure reliability.