PenguinHarness
An open-source, local-first platform that automates building, evaluating, optimizing and deploying multi-agent AI applications, generating a runnable agent app from a single natural-language description and having agents evaluate and rewrite their own skills.
PenguinHarness runs on a user's own computer or server and automates the lifecycle of an agent application, from initial creation through evaluation, optimization and deployment, rather than requiring an agent to be assembled by hand with a library such as LangChain. A single natural-language description, for example asking for a documentation-answering RAG app built from a given GitHub repository, produces a complete, runnable scaffold with working code and run instructions.
Its toolset is deliberately minimal, favoring low-level interfaces with fewer tool calls and tokens, and is tuned for open models such as DeepSeek, aiming for lower per-task cost than a general-purpose coding harness on the same benchmark tasks.
A self-evolution mechanism lets an agent run a benchmark against its own skills, identify where it lost points, and ship an improved version, keeping a snapshot before each round and making every request in the process observable in a trace view. Built-in plugin categories cover office productivity tasks such as data analysis and web scraping, software development, and AI app development, and agents can also write and optimize their own additional skills.