PenguinHarness

PenguinHarness builds runnable AI agent applications from plain-language requirements on your computer or server. It's for developers who want agents to handle app creation, evaluation, tuning and deployment.

• Agent app creation. Describe an application and have an agent produce its scaffold, code and run instructions, with a CLI and SDK available for scripted work.
• Evaluation and tuning. Parallel evaluators score agents against benchmarks, and an optimizer uses scores and execution traces to revise prompts and Skills; snapshots let you restore an earlier state.
• Agent workflows. Manage agents and separate chat sessions, schedule unattended tasks, and delegate work to isolated subagents, with built-in Skills for browser automation, data analysis and software development.
• Trace and approval controls. Inspect model requests and tool calls, including failures, timing and token use, and review tool calls through approval checks.

The desktop app runs on macOS, Windows and Linux. A Docker server provides the web interface on your own infrastructure. PenguinHarness is open source under Apache 2.0 and supports local models and cloud providers through OpenAI-compatible endpoints, including DeepSeek, OpenAI, Anthropic and OpenRouter. The app stores data locally; cloud model requests go to the selected provider. Tuning stays within the agent's workspace and Skills, leaving the harness core unchanged.