Platform overview
Constellation OS is a control plane for heterogeneous space assets. It gives operators one place to see, question, and reason about a fleet regardless of who built the satellites, who runs the gateways, or where the telemetry originates: one telemetry contract in, one fleet graph out, predictions and deterministic analysis on top, and an operator console that treats every asset the same way.
The design premise is that space is becoming a software-defined, multi-operator network, and the durable layer is the one that coordinates the physical assets: what is allowed, what is happening, what will fail, what should change, and who is accountable. Constellation OS is built to be that layer, which is why tenancy, provenance, auditability, and determinism show up as product features rather than implementation details.
Components
| Component | What it is | Where it runs | Docs |
|---|---|---|---|
| Console | Globe-first operator surface: timeline, entity cards, assistant, reports | Browser SPA served from CloudFront | Console |
| Assistant (console agent) | Natural-language surface that routes questions to deterministic tools and prediction models | In the console, backed by a chat service | Agentic |
| Platform API | Tenant-scoped HTTP surface: telemetry ingest, topology reads, predictions, health | Per-environment API service | API |
| Fleet agent | Edge CLI that posts telemetry and pulls topology from gateways and ground sites | Your infrastructure | Integration bundle |
| Prediction models | Trained model families served per tenant; SNR is the production family today | Platform-side inference | Predictions API |
| Deterministic engines | Fixed-step, seedless-randomness-free analysis engines: fleet statistics, coverage, link budgets, outage and rain scenarios, optimizers | In the console, against the live scene | Simulations and scenarios |
| Per-tenant data planes | A dedicated time-series database instance and encryption key per tenant | Platform infrastructure | Tenant isolation |
How the pieces connect
- Ingest: your pipeline or the fleet agent posts telemetry records to the platform API. Every write is stamped with your tenant and lands in your dedicated data plane.
- Read: the topology endpoint projects the latest-known state per entity out of that telemetry, which is what the console and your integrations poll.
- Predict: the predictions endpoint serves cached or live SNR forecasts per link, with full provenance on the live path.
- Operate: the console renders the fleet on the globe against a single simulation clock, and the assistant answers questions by running deterministic engines and prediction models over the same scene the operator is watching.
- Report: analyses capture into reports with their method, assumptions, and provenance attached, so conclusions survive review.
What the platform is not
Honest boundaries, so you can plan around them:
- It is not a traffic-routing or network-orchestration layer; it is the fleet operations layer beside one.
- The API has no webhooks, no streaming, and no pagination; integrations poll. See API overview.
- One prediction model family (
snr) is in production; other prediction domains in the console are clearly labeled baselines or stubs until their backends ship.
Where next
- Concepts: tenant, fleet, the unified asset model, and the honesty principles.
- Data model: the record shapes that flow through the system.
- Architecture: how the platform and console are actually built.
- Security and Enterprise: isolation, keys, and dedicated stacks.