How Scry works

Scry has two parts: the app on your phone and a small server, scry-connect, on your robot. They talk directly over your network — there is no Scry cloud in between.

flowchart LR
    A["Scry app\non your phone"]
    B["scry-connect\non your robot"]
    C["Your ROS 2 graph\ntopics · nodes · services"]
    A <-->|"your network"| B
    B <-->|"ROS 2"| C
    classDef brand fill:#292826,stroke:#3A3835,stroke-width:1px,color:#E8E4D9
    class A,B,C brand
    linkStyle 0,1 stroke:#A3B86C,stroke-width:2px,color:#9C9A8D

The phone does the thinking

Your phone is the brain. It runs the Scry assistant, decides which actions to take on the robot, renders the results as readable panels and charts, keeps your background monitors running, and manages your fleet of robots. The AI provider you choose (see Choose your AI) powers the assistant; your phone talks to it directly.

The robot just exposes its ROS 2 graph

scry-connect is a lightweight server you run on the robot. It makes the robot's live ROS 2 graph — topics, nodes, services, parameters, lifecycle, diagnostics, transforms, and more — available to Scry as a set of safe, well-defined actions. It uses the standard ROS 2 client libraries, so it works with any ROS 2 middleware (Fast-DDS, CycloneDDS, Zenoh, Connext, and others). You don't configure any of that — Scry follows whatever your robot already uses.

What happens when you ask a question

1. You ask Scry something in plain language — by text, voice, or image.
2. Scry figures out which information it needs and asks scry-connect for it over your network.
3. The robot returns the data — a topic's rate, a node's connections, a diagnostic summary, a camera frame.
4. Scry reads the results, explains what they mean, and renders them as panels, plots, or tables instead of raw data.

Scry can chain several of these steps in one answer — for example, reading a topic, then checking a related parameter — before giving you a single, clear explanation.

Reads are free; actions ask first

Scry separates two kinds of capability:

• Reading the robot — listing topics, inspecting nodes, checking diagnostics — happens freely. Nothing changes on the robot.
• Acting on the robot — publishing to a topic, setting a parameter, calling a service, triggering a lifecycle change — always asks for your tap-to-approve first. Scry shows you exactly what it intends to do, and nothing is sent until you approve it.

This is the core safety model: the assistant can suggest an action, but you are always the one who authorizes it.

No cloud backend

Everything runs between your phone, your AI provider, and your robot. There is no Scry server collecting your data, no telemetry, and no account required to connect to a robot on your own network. For a full breakdown of what is stored and where, see the Privacy policy.

For the full list of what Scry can inspect and do on a robot, see What Scry can do.