Interaction System Card
This system card follows the Anthropic / OpenAI System Card format to fully disclose Interaction's capability boundaries, safety design, privacy guarantees, and known limitations.
This system card will be updated with each product iteration. To report gaps or inaccuracies, contact us at contact@lumirain.cn.
Product capabilities: multi-model chat (Claude / GPT / Gemini, etc.), local VM task execution, Code Mode SDK integration. Not supported: real-time internet access (except within Task Mode sandbox), Computer Use, AI companion features.
Safety design: task execution runs in an isolated VM with no default access to the host filesystem; users precisely control file authorization scope; all actions are reversible via snapshot mechanism.
Known limitations: complex multi-step task success rate is bounded by context window constraints; Chinese code comment recognition accuracy is lower than English; voice input accuracy depends on device hardware.