The model, without the stack.
License X1D inference for your own product.
Retrofit Studio is a full deployment path — perception, enforcement, runtime, OEM integration, evidence records, commissioning. That is the right answer when you need all of it.
It is not the right answer when you already have hardware control, your own safety layer, or your own deployment infrastructure, and you need a foundation model that understands physical environments.
The Xolver VLA SDK gives you the inference boundary. Nothing more.
What you get
A signed license key activates a self-hosted policy server. The server accepts observations over WebSocket, runs X1D inference, and returns actions. You control the hardware. You control the safety boundary. We provide the model.
$ pip install xolvervla-client
from xolvervla_client import WebsocketClientPolicy
policy = WebsocketClientPolicy(
host="your-inference-host",
port=8000,
api_key="xvk-..."
)
action, state = policy.infer(observation)
The client handles reconnection, version negotiation, and typed errors. Your application handles everything else.
Most teams go from pip install to a first inference call in under an hour once a license is issued — there is little glue code to write.
What the SDK includes
- X1D inference over WebSocket
- Signed license keys with configurable connection limits
- Prometheus metrics endpoint at /metrics
- Health endpoint at /healthz
- Lightweight Python client with auto-reconnect and backoff
- Deployment modes: development, production, certified
What the SDK does not include
- Retrofit Studio
- OEM hardware adapters
- Safety shield enforcement
- Industrial templates
If you need those, use the full platform.
Who this is for
OEM controller manufacturers
You make the arm. You ship the controller software. You need a foundation model your customers can activate by license rather than integrating a new stack.
Robotics platform companies
You already have hardware abstraction, safety logic, and a deployment pipeline. You need X1D at the inference boundary, not a second deployment system running beside yours.
Research and evaluation teams
You need a production-grade VLA model to benchmark against, test integration assumptions, or run comparative studies. You need inference access, not an industrial deployment workflow.
Supported OEM arms
The SDK is OEM-agnostic at the inference boundary. If your controller talks to one of these arms, it talks to Xolver.
KUKA arms connect to Xolver through a dedicated compatibility pack — retrofit your cell through Retrofit Studio, or skip the retrofit path entirely and embed X1D inference directly through the SDK above.
License structure
License keys are cryptographically signed tokens that encode connection limits, OEM pack entitlements, and expiry. Validation happens locally at connection time, so the policy server keeps running even without a live connection to a license server.
Decodes and prints the license without connecting to anything.
FAQ
Do I need Retrofit Studio to use the SDK?
No. The SDK is specifically designed for teams that already have their own deployment infrastructure. You run the policy server, connect with the Python client, and handle hardware control yourself.
What does "self-hosted" mean?
The policy server runs in your environment — on-prem, on a cloud VM, or on an edge device. Xolver provides the server binary and the license key. You operate it.
Can multiple machines share one license?
Connection limits are encoded in the license key. Enterprise and OEM licenses support multiple simultaneous connections. Contact us to discuss your scale.
What safety guarantees does the SDK provide?
The SDK provides X1D inference. Safety enforcement is your responsibility — by design. If you need Xolver's safety shield and enforcement layer, use Retrofit Studio instead.
Request SDK access
Tell us about your platform, your target deployment environment, and what you need X1D to do. We'll follow up with license options and a technical brief. Once approved, expect your first inference call the same day — most of the integration work is the few lines of code shown above.
hello@xolver.ai