Developer builds self-coding desktop AI agent with human approval gates
A developer has released AMA-teras, an open-source desktop AI agent built with Electron and TypeScript that can write its own tool plugins when it encounters a capability gap. The agent supports swappable AI models from providers like Anthropic, OpenAI, and Moonshot, and runs entirely on the user's local machine. Every piece of generated code must pass typechecking, unit tests, and a smoke run before a human reviews and approves the full diff via a git-based pipeline. The system includes protections against self-modification of its core approval mechanism, and community plugins require verified audit evidence given that studies suggest 12–20% of marketplace skills can be malicious. Currently available as a Windows installer, the project is licensed under AGPL and the developer is seeking feedback on its safety model.
This is an AI-generated summary. ShortSingh links to the original source for the complete article.
Discussion (0)
Log in to join the discussion and vote.
Log in