The latest ARMOR update adds an embedded help system with an AI-powered assistant. The help pages mirror exactly what is on this website, but they are compiled into the app build, so they are available offline. The assistant uses Apple Intelligence on iOS 26 and later, so it runs locally: none of your questions or data leave your device.
Running a local model means working inside a small context window, without the encyclopedia-sized windows of cloud assistants. ARMOR handles that with an adapted form of Retrieval Augmented Generation (RAG):
- The whole documentation set lives in one JSON file, in several languages, split into chunks.
- At run time the app computes an embedding for each chunk.
- When you ask a question, it combines semantic and lexical search to find the chunks that match best.
- The assistant then reads and summarizes those chunks before replying.
The assistant can also read a long URDF by breaking it into a model tree. A question about the robot’s shoulder then gets an answer about the whole pitch, roll, and yaw series of three revolute joints, not just a single joint.
Key Points
- Help is bundled in the app, so it works offline on iPhone, iPad, and Mac
- The assistant runs on device with Apple Intelligence (iOS or macOS 26 and later), with no data sent off the device
- Retrieval combines semantic and lexical search over documentation chunks, in multiple languages
- Questions about a loaded robot are answered from a model tree built from its URDF
- The design behind the assistant is explained in the ARMOR v1.2 release post
See Built-In Help and Ask for how Browse, Ask, and Videos work in the app, and Robot Simulation on iPhone for ARMOR on the phone.