Sidekick Robotics introduced the Builder beta on October 23, a developer plan for connecting AI models to robotic hardware. The announced price is $7 a week after a free first week. This provides access to a development platform; it does not purchase a robot or demonstrate that an autonomous production cell is ready.
The company describes a catalogue with different access states, a primary-model selection and a route preview before execution. The important distinction is between a model appearing in a catalogue and a service that can actually run. A research label does not mean the project can immediately control a connected device.
One interface, different meanings of a result
Earlier API documentation separates predicted frames, spatial coordinates, robotic actions and evaluation. It also documents a field identifying simulated placeholder responses. Successfully receiving JSON therefore does not establish that a model actually ran or that its output is a suitable command for motors.
This separation has practical consequences. An answer saying where an object is describes an observation; it does not automatically define a permissible movement towards it. An integration needs to track the transformation between camera coordinates, robot coordinates and the control command. Checking that contract remains part of the project even when external requests share one format.
Our editorial assessment is that a platform of this kind should be evaluated through a reproducible task, rather than the number of models in a menu. A useful pilot would retain the precise version, input observations, selected route and reason for any provider fallback. A failure can then be examined as a specific record instead of remaining a general impression from a demonstration.
The subscription does not describe the entire bill
The pricing page distinguishes Builder from Enterprise and the forthcoming Team plan. The API announcement describes separate usage units, including actions and predicted frames. A weekly subscription should not be interpreted as a promise of unlimited free execution of every external model.
When comparing offers, a team should request the complete cost of a defined test: preparation, calls, retries after failures and required hardware. This allows a seemingly inexpensive request to be compared with a task actually completed. SIExpo has not opened a paid account or checked the service's invoices or performance.
Software selection does not replace machine limits
Sidekick describes an approach with constraints below the model layer, responses to hardware signals and human involvement during learning. This is the system developer's stated approach, rather than independent certification of any robot connected to it.
For a first pilot, we recommend a controlled space, an explicit stop mechanism and predetermined behaviour when results are missing or late. Testing should also cover lost connectivity, model switching and incompatible formats. None of those situations should be hidden inside an apparently successful execution.
What Builder adds is a more accessible organisational framework for these experiments. Whether it reduces integration work for a particular team must be established on that team's hardware and tasks. The announced features are Sidekick's descriptions; this article does not report an independent test on a real robot.
*Cover: a conceptual AI illustration.*
*Sources checked: 2026-10-23T16:10:40.049366+00:00.*
