Developer API

Spatial intelligence
for hardware AI.

One API to give your drones, robots, and autonomous systems the ability to reconstruct, understand, and simulate the physical world in real time.

Quick Start →Get API Key

Python · TypeScript · REST · gRPC · 99.9% uptime SLA

Quick Start

From sensor to spatial intelligence in 10 lines.

Our Python SDK handles the full pipeline — reconstruct from any source, build a queryable scene graph, and run physics simulations. All with a single API key.

[python][typescript][rest-api][grpc]

3endpoints

1SDK

<5minto first call

pipeline.py

import percept

# Initialize with your API key
client = percept.Client(
  api_key="pk_live_..."
)

# Reconstruct from drone video
scene = client.reconstruct(
  source="drone_capture.mp4",
  mode="metric_3d"
)

# Query the scene graph
defects = scene.query(
  type="structural_defect",
  severity="critical"
)

# Simulate physics
result = scene.simulate(
  scenario="fire_spread",
  duration=4,
  wind_speed=12.5
)

# → 2.4M vertices · 12 defects · 96.2% confidence

Endpoints

Three endpoints. The entire pipeline.

Raw sensor input goes in. Structured spatial intelligence comes out. Each endpoint builds on the last.

Reconstruct

percept.reconstruct()

Turn any video, LiDAR scan, or satellite image into metrically accurate 3D. Supports phone cameras, drones, terrestrial scanners, and satellite.

Param	Type	Description
source	str	Video file, LiDAR point cloud, or URL
mode	str	"metric_3d" \| "mesh" \| "point_cloud"
coordinate_system	str	CRS code (default EPSG:4326)
resolution	float	Target resolution in meters

percept.reconstruct()

scene = percept.reconstruct(
  source="drone_capture.mp4",
  mode="metric_3d",
  coordinate_system="EPSG:4326"
)

# → 2.4M vertices · ±2cm accuracy · 3m 12s

Query

scene.query()

Query the physical world like a database. Filter by object type, material, defect severity, spatial region, or semantic label. Returns structured results with coordinates.

Param	Type	Description
type	str	Object or defect type to find
severity	str	"critical" \| "high" \| "medium" \| "low"
within	Bounds	Spatial filter (bbox, radius, polygon)
material	str	Filter by detected material type

scene.query()

defects = scene.query(
  type="structural_defect",
  severity="critical",
  within=scene.bounds(
    lat=37.87, lon=-122.26, radius=500
  )
)

# → 12 critical defects · georeferenced

Simulate

scene.simulate()

Run real physics on real geometry. Fire spread, structural failure, flood inundation, corrosion — all grounded in the actual reconstructed environment.

Param	Type	Description
scenario	str	"fire_spread" \| "structural_failure" \| "flood" \| "corrosion"
duration	int	Simulation hours
wind_speed	float	Wind speed in km/h
wind_direction	int	Wind bearing in degrees

scene.simulate()

result = scene.simulate(
  scenario="fire_spread",
  duration=4,
  wind_speed=12.5,
  wind_direction=225
)

for frame in result.frames():
    print(f"t={frame.time}s  area={frame.area_m2}m²")

# → 96.2% prediction accuracy

SDK

Everything else you need.

Beyond the three core endpoints — streaming, export, edge deployment, versioning, and event-driven workflows.

Streaming Ingest

percept.stream(feed, fps=30)

Real-time processing of live sensor feeds. Continuously update the scene graph as new data arrives.

Export & Integrate

scene.export("gltf", crs="EPSG:4326")

Export to glTF, LAS, GeoJSON, or OBJ. Integrate with GIS, game engines, CAD, or your own pipeline.

Webhooks & Events

percept.on("defect_detected", cb)

Event-driven architecture. Trigger workflows when objects change, defects appear, or thresholds breach.

Batch Processing

percept.batch(sources=[...], parallel=4)

Process thousands of captures in parallel. Automatic queuing, retry, and progress tracking built in.

Edge Deployment

percept.edge(device="jetson", model="v3")

Run reconstruction on NVIDIA Jetson, Qualcomm, or custom edge hardware. 30fps on-device inference.

Version Control

scene.checkpoint("pre-storm-2024")

Snapshot and diff scene graphs over time. Track how environments change between captures.

Spatial Memory

objectinfrahazardfleet

memory.find(sku="4421", max_age_s=30)→ aisle A · 1.2s ago · robot 1 observed

memory.recall("charging bay")→ [13.2, 23.1, 0.4] · conf 0.99

fleet.sync(agents="all")→ 3 agents · 284 objects synced · 247ms

SEMANTIC MAP v3284 objects indexed2 agents · live sync

Every robot.
One shared memory.

Percept gives your fleet a persistent semantic 3D map of every environment they encounter — every object indexed by type, location, and last seen time. When one robot discovers something, every robot knows.

Warehouse fleets: any picker instantly knows where any SKU was last seen — even if another robot moved it
Home robots that remember where you left your keys, your glasses, your phone
Search & rescue teams building a live shared map of survivors, hazards, and cleared zones
Query the memory graph by object type, confidence, recency, or spatial region via API

fleet_memory.py

from percept import memory, fleet

# Any robot queries shared fleet memory
shelf = memory.find(
  object="package", sku="4421",
  max_age_s=30
)
# → aisle A · 1.2s ago · robot 1 observed

# Sync memory map across entire fleet
fleet.share(
  map=memory.export(),
  agents="all"
)
# → 3 agents · 284 objects synced · 247ms

Get Started

Start building in minutes.

Get API Key →Explore SpatialOS

Free tier available · No credit card required · Full SDK access

Spatial intelligencefor hardware AI.