AlterEgo: Towards the Zero-Cost Personal Agent - open-source Windows-native agent at $0 model cost

A preprint titled "AlterEgo: Towards the Zero-Cost Personal Agent" (DOI 10.5281/zenodo.23446166) and its accompanying open-source Windows-native code, openmuse-win, were published, presenting a failover router that multiplexes 4 free-tier LLM providers — NVIDIA, ModelScope, Zhipu and Agnes — with per-role binding and automatic fallback, for total model spend of $0 during development. The authors argue zero marginal model cost should be a first-class design goal for personal agents and include a taxonomy of personal-agent cost models covering subscription, paid-default, BYOK, allowance, free-tier failover and self-host. The agent executes on Windows via UI Automation, screenshots and a persistent Chromium rather than Docker sandboxes or cloud VMs; the project is described as early-stage and seeking contributors.

Hi all, We just published a preprint and open-sourced our personal agent project. Wanted to share with the research community: Paper: AlterEgo: Towards the Zero-Cost Personal Agent - DOI: AlterEgo: Towards the Zero-Cost Personal Agent | Zenodo https://doi.org/10.5281/zenodo.23446166 Code: GitHub - wojiadexiaoming-copy/openmuse-win: Windows本地智能体 · GitHub https://github.com/wojiadexiaoming-copy/openmuse-win Core idea: A personal agent that cannot afford to stay on is not a personal agent. We argue zero marginal model cost should be a first-class design goal, and show an existence proof: a failover router multiplexing 4 free-tier LLM providers NVIDIA, ModelScope, Zhipu, Agnes with per-role binding and automatic fallback. Total model spend in development: $0. The paper also includes a taxonomy of personal-agent cost models subscription / paid-default / BYOK / allowance / free-tier failover / self-host - the first we’re aware of. On the execution side: Windows-native UI Automation, screenshots, persistent Chromium rather than Docker sandboxes or cloud VMs. Happy to answer questions or receive feedback. The project is at an early stage and we welcome contributors.