today, xiaomi’s mimo team officially launched and open-sourced a brand-new ai programming assistant for end devices—mimo code v0.1.0. as an exploratory tool deeply optimized for developers, it has been rebuilt and upgraded based on the opencode project, adheres to the permissive mit open-source license, and comes pre-integrated with the self-developed multi-modal model mimo‑v2.5, available free of charge for a limited time. it also supports mainstream large models such as deepseek, kimi, and glm, as well as third-party token services, balancing flexibility and extensibility.
its core breakthrough lies in pioneering a “triple persistent memory architecture”: integrating project-level contextual memory, session checkpoint snapshots, and task progress tracking, thereby fundamentally addressing the common issue of information decay in long‑range interactions. even across hundreds of rounds of dialogue, critical logic and state remain accurately preserved. the design employs a master–slave separation mechanism: the main agent focuses on reasoning and execution, while independent sub‑agents asynchronously handle state archiving and summary generation. when the context window approaches saturation, the system automatically generates a concise session summary, ensuring the main workflow remains lightweight and efficient.
to maximize the performance of the mimo series models, the project has built a dedicated harness runtime framework and seamlessly integrated the compose intelligent collaboration mode. users simply press tab to enter this mode and input their initial ideas, triggering an end-to-end automated pipeline that covers requirement analysis, architectural design, code generation, unit testing, and quality review, ultimately delivering industrial‑grade code ready for production.
in terms of user experience, mimo code natively supports voice control. leveraging the high‑precision mimo‑v2.5‑asr speech recognition module, it enables key operations such as natural language command correction and hands‑free “send” and “execute” functions. installation is compatible across multiple platforms: mac and linux users can deploy with a single curl command, while windows users are recommended to install via npm for quick setup.
