Use Scry¶
How to actually drive the app day-to-day, once you're past first install.
- Chat with the agent — the primary surface. Streaming responses, inline tool result rendering, multi-turn context, retry, fork, edit.
- File attachments and images — drop a log file or a screenshot into the conversation. Up to 200 KB text, max 2048 px images (auto-downscaled).
- Voice input — Android
SpeechRecognizerintegration. No audio leaves the device. - Background monitors — edge-triggered alerts when a threshold trips. Fires into chat as if the assistant posted it.
- Sending feedback — inline thumbs-up / thumbs-down on every assistant reply, plus a top-level form for general impressions, bug reports, and feature requests.
The five tabs¶
| Tab | What's there |
|---|---|
| Fleets | List of paired robots. Add/remove, switch active. |
| Robot | Dashboard for the active robot — health, nodes, topics summary. Honest data sheet, no fabricated verdicts. |
| Scry | Primary chat surface. Where you spend 90% of your time. |
| ROS | Browse hub — 10 ROS entity families (topics, nodes, services, actions, lifecycle, params, components, logs, TF, processes). |
| Viz | Long-running visualization — scene, camera, BT, geomap, plot, sensors, teleop. |
The Scry tab is the center button on purpose. Chat is the primary interaction model; everything else supports it.
What Scry isn't¶
- Not Foxglove. No comprehensive visualization layer, no recording playback, no PlotJuggler-grade timeseries. The Viz tab covers the common cases; for deep analysis, use the right tool.
- Not a remote desktop. You can't run arbitrary commands on the robot — only the ~99 MCP tools the connect exposes. (You can extend the connect with custom tools if you need to.)
- Not autonomous. The agent proposes; you approve. Every "write" tool call (publish to a topic, set a parameter, call a service) requires an explicit tap before it dispatches.