SpatialAI SDK

The CUDA of spatial sensing

One SDK for every VisionLibra device — capture, calibrate, run models, and hand events to agents. Python for speed of thought, C++ for production, ROS 2 for robots.

Free Quickstart
Developer Experience

From unboxing to point cloud in minutes

Auto-discovery, unified device API, and prebuilt wheels for x86, ARM, and Jetson. The same script runs on a $9 sensor and a $99 robotics LiDAR.

  • Python API shipping today (v0.3 on PyPI) — C++ and ROS 2 on the roadmap
  • Depth-CV models included free: presence-core, occupancy-home-v1
  • Simulator for every device — develop before hardware arrives
  • Real hardware backends: Spatial Mini (I²C) and Spatial Home (UART, full protocol)
  • OTA client, fleet enrollment, and point-cloud streams in development
pipeline.py
from spatialai import Camera, Pipeline

cam = Camera()                       # auto-discovers any VisionLibra device
pipe = Pipeline(cam)

pipe.load_model("people-tracking-v3")  # one line from the marketplace
pipe.enable("point_cloud", "heatmap")

for frame in pipe.stream():
    print(frame.people, frame.fps)     # 30 fps, all on-device
Architecture

Where the SDK sits

SensorEdge RuntimeSpatialAI SDKModelsAgentsFleet Cloud
🧠

Edge Runtime

Schedules capture, inference, and agent rules on-device — deterministic latency, no cloud dependency.

📦

Model Loader

Pull marketplace models with load_model() — quantized and tuned per device automatically.

☁️

Fleet Client

Enroll a device once; OTA updates, logs, and heatmaps flow to the dashboard from then on.

Platform Support

Where it runs today

Honest status, updated with every release — see the changelog.

PlatformPython APISimulatorSpatial Mini (I²C)Spatial Home (UART)C++ / ROS 2
Linux x86_64✅ v0.3✅ with I²C adapter✅ beta, USB-serial🚧 roadmap
Raspberry Pi (ARM64)✅ v0.3✅ native I²C✅ beta, native UART🚧 roadmap
NVIDIA Jetson✅ v0.3✅ native I²C✅ beta, native UART🚧 roadmap
macOS✅ v0.3— no I²C bus✅ beta, USB-serial🚧 roadmap
Windows✅ v0.3— no I²C bus✅ beta, USB-serial🚧 roadmap

Python ≥ 3.9. Hardware backends are datasheet-accurate with protocol-level test coverage (beta until validated on production modules). Spatial Vision ships as a decode/filter toolkit (spatialai.opn6001) — full V4L2 capture needs platform MIPI bring-up; Spatial Robot (UVC) is in development.

Licensing

Start free, license when you ship

Community
$0
  • Full Python API
  • Evaluation & prototyping
  • Community forum support
Start Free
Enterprise
$5,000 / year
  • Everything in Pro
  • Agent platform & connectors
  • Dedicated engineer