Agent-harness – A minimal, composable Go library to build AI Agent Harnesses (github.com)

A new minimal, composable Go library named **Agent-harness** has been released, designed to facilitate the creation of agentic tool-calling loops that operate on top of large language model (LLM) APIs. Compatible with Go 1.26 and higher, this library enables developers to implement a streamlined agent lifecycle: calling the LLM, executing specified tool calls, and recursively feeding results back into the loop. Users can customize aspects such as system prompts, message handling, and tool execution while managing conversation storage independently. The significance of Agent-harness lies in its adaptable architecture that allows developers to focus on high-level agent behaviors rather than low-level implementation details. With built-in functionality for handling multiple provider adapters (such as OpenAI and Anthropic), pause/resume mechanisms, and hooks for event handling and observability, the library supports the orchestration of complex workflows, including multi-agent interactions. Its comprehensive documentation, including architecture and provider integration details, empowers developers to create sophisticated AI-driven applications with ease, marking an important step forward in the AI/ML community's ongoing evolution of automated, intelligent agents.