Orbital Runtime Live
Constraint-Aware
Execution
The first API for orbital compute scheduling. POST a satellite ID and a workload, get back a physically-accurate execution plan — computed from real orbital mechanics.
The Problem
Satellites generate 1 TB/day. A ground station pass transfers ~7.5 GB. That's a 130:1 mismatch.
Computation must be split between space and ground. Some steps reduce data 190:1 and belong on-board. Others need ground resources. Data crosses the space-ground boundary through narrow contact windows with varying link quality. Nobody offered an API that solves this. Until now.
Four-Phase Planning
Every plan is computed from real orbital mechanics — SGP4 propagation, eclipse detection, ground station passes, and link budgets.
Environment
SGP4 propagation at 30-second steps. Eclipse detection via cylindrical shadow model. Ground station passes for 12 stations. Link budget per pass.
Placement
Compare on-board cost (energy + thermal + time) vs ground cost (transfer overhead). Steps with >10:1 data reduction are forced on-board.
Transfers
Auto-insert downlink/uplink at space-ground boundaries. FEC selection, encryption overhead, multi-pass allocation for large payloads.
Scheduling
On-board steps in orbital windows. Transfers pinned to pass windows. Ground steps run on arrival. Dependencies resolved across boundaries.
POST /v1/plan
Content-Type: application/json
{
"satellite_id": "25544",
"preset_id": "split-learning"
}
Response:
{
"placement_decisions": [
{ "step": "feature_extraction",
"location": "onboard",
"reason": "40:1 data reduction" },
{ "step": "train_backend",
"location": "ground",
"reason": "requires persistent storage" }
],
"transfer_schedule": {
"total_downlink_mb": 52.82,
"total_uplink_mb": 7.55
},
"error_budget": {
"delivery_confidence": 0.99,
"fec_overhead_mb": 14.0
}
}Five Workload Presets
Real orbital compute patterns — from on-board inference to distributed training to error-resilient relay.
On-Board ML Inference
2 GB capture, 10.5 MB downlink. 190:1 data reduction. All compute stays on the satellite.
4 steps · All on-board · Minimal downlink
Split Learning
Feature extraction on-board (layers 1-3), training on ground. 36.75 MB down, 5.25 MB up. Bidirectional transfer scheduling.
8 steps · Hybrid · Bidirectional
Earth Observation QA
5 GB raw imagery through on-board QA, cloud filtering, FEC encoding. 560 MB downlink across multiple passes.
8 steps · Multi-pass · QA filtering
Federated Learning
Only gradients leave the satellite — never raw data. Privacy-preserving by design. 3.7 MB down, 5.8 MB up.
9 steps · Privacy-first · Tiny transfer
Resilient Store-Forward
Reed-Solomon erasure coding (rate 2/3). Any 2-of-3 blocks can reconstruct. Two separate pass windows.
5 steps · Erasure coded · Fault-tolerant
Custom Workloads
Define your own step DAG via the API. The planner handles placement, transfers, and scheduling for any workload topology.
Coming soon
Real Physics. Not Toy Models.
SGP4 Orbital Propagation
Full SGP4/SDP4 via satellite.js. 30-second timesteps, 12-hour prediction window. The same algorithm NORAD uses for orbital prediction.
Eclipse Detection
Cylindrical shadow model computing sun-earth-satellite geometry. Accurate to ~10 seconds — sufficient for energy-aware scheduling.
12-Station Ground Network
Real pass prediction for Svalbard, Fairbanks, Santiago, Awarua, and 8 more. Elevation-dependent link budgets from 25 to 120 Mbps (X-band).
Error Correction
Adaptive FEC selection (rate 1/2 to 7/8) based on channel BER. Retransmission reserves. Deterministic delivery confidence calculation.
Security Overhead
AES-256 encryption, SHA-256 integrity checking, key exchange timing. All overhead factored into transfer scheduling and pass allocation.
Deterministic Plans
Same inputs always produce the same plan. BER-based fault model, not random. Reproducible for testing and validation.
API Endpoints
GET /v1/presets # List workload presets
POST /v1/plan # Create execution plan
GET /v1/plan/:id # Retrieve a plan
GET /v1/plan/:id/events # Execution event stream
# Base URL
https://rotastellar-cae.subhadip-mitra.workers.dev
# No API key required for CORS-allowed origins
# Plans are stored for 1 hourGet Started
See it in action
Open the satellite tracker, select any satellite, and run an execution plan from the Schedule tab.