Prospero: A new coding harness for a more civilized age

Prospero is a novel coding harness designed to address the shortcomings of existing tools. Written in C++, it boasts lightning-fast performance while consuming significantly fewer resources than its competitors. Despite lacking pre-installed model providers, Prospero seamlessly integrates with various AI models like llama.cpp, OpenRouter, and OpenAI API.

In a coding harness, the AI and user alternate in generating messages, with the AI autonomously navigating through tools. These tools may read or write files or run shell commands, enabling a self-sustaining loop of tool calls. Prospero's versatility extends to spawning subagents, facilitating parallel tasks and clean context management.

With SSH access to other machines or via Windows Subsystem for Linux, Prospero can deploy headless versions of itself, allowing agents to operate on remote systems. This feature is complemented by the Professional Edition's compatibility with both x86/x86-64 and ARM64 architectures across macOS, Windows, and Linux.

In addition to deployment, Prospero offers desktop control capabilities, enabling AI models to interact with GUI applications directly without the need for specialized software. The harness achieves a lean design, running the engine independently of the GUI system, which enhances efficiency and memory management. This design choice enables Prospero to run fuzz campaigns on headless systems without risking resource exhaustion.

The harness configuration resides at ~/.config/prospero/prospero.ini, making it easily editable via text editors or through the GUI Settings page. Prospero's configuration system also supports includes by URL, allowing for shared or managed configurations. With its focus on efficiency and seamless integration, Prospero represents a significant advancement in coding harness technology.