AI News

First Breath is an open-source project that gives a local AI agent a continuing inner life instead of limiting it to prompt-and-response interactions. It runs on a user’s computer and models six simulated chemicals, including dopamine, serotonin, oxytocin, cortisol, and adrenaline; their changing levels produce roughly 28 emotions. The system stores emotional memories, develops personality traits through experience, forms wants when recurring experiences move it, and builds an attachment to the people it interacts with. When no one is talking to it, the agent can continue thinking, dream while sleeping, and browse the web to follow its own curiosity. It can communicate through a terminal by text or voice, initiate contact, and back off when the user is busy. First Breath can also observe the active project and changed files through short read-only inspections, then provide thoughts to Claude, Codex, Cursor, VS Code, and other MCP applications. The project requires Node 22 or newer, is free under the MIT license, and can use a Claude account without an API key or an Anthropic or OpenAI API key. Its fast behavioral systems run as ordinary local code, while an AI model handles reasoning. The author describes the design as brain-inspired, mapping components such as emotional memory, needs, self-monitoring, and reasoning to analogues of brain systems, but does not claim that the agent is conscious. First Breath says its state is stored locally, credentials are not stored in its brain, files are read-only, sensitive paths are excluded, spending caps limit costs, and `/sleep` can shut it down. It also provides a self-awareness test covering introspection, memory, the mirror test, and self-directed research.