Spaceintelligenceforall.
Intelligence is the next frontier of space. The first public release of the OS Platform is making it accessible to everyone.
By Omeed Tehrani and Raaid Kabir
The Constellation Platform is live! Anyone with an account can now use the forecasts we run for satellite operators. The Platform predicts link quality and traffic demand for the fleets and ground networks you operate or study.
While we built this release, we worked rigorously with our research team and studied how the companies behind the biggest platforms in computing and AI grew ecosystems around their technology. The same lesson showed up every time we looked. They won by making it easy for other people to build on what they made, and by creating strong feedback loops so that every new builder made the platform better. We want to do that for space, with operators, integrators, students, and researchers using our frontier intelligence to build the tools that put alarm-driven mission operations behind us and move space toward autonomous operations.
This first release has two model families, an API and agent tools to use them, and an academic program that gives students and researchers the same access as a paying developer.
Why forecasts
Most satellite operators still find out about a problem when an alarm fires, and then a person works out what to do. That approach is reaching its limit. About 11,000 active satellites orbit Earth today, with projected growth targeting between 70,000 and 100,000 by 2030, and even highly automated constellation operators say that operations built around people in the loop do not scale to hundreds of satellites.
Weather makes the problem harder. Operators plan contacts in advance and size their weather margins from long-term statistics, but as links move to Ka band and optical, a single storm cell or cloud can decide whether a pass delivers any data. A pass booked under a clear sky can come back empty if a storm moves in before it starts. A forecast of the link tells an operator how much margin a pass actually needs, or which station to book instead. Our post on physics-informed contact planning shows how much more data that moves for every booked minute of antenna time.
The missions we are building for
We support research, commercial, and government missions—and the ground stations that connect them to Earth. From a team's first CubeSat to an operational constellation, every mission depends on making the most of its communications links.
- Communications satellites and constellations
- Operators delivering connectivity across low, medium, and geostationary Earth orbit, where link availability, gateway coverage, and handovers determine how much capacity reaches customers.
- Earth observation and remote sensing
- Missions collecting imagery, weather observations, and environmental data, often through shared ground station networks. Every downlink opportunity matters when a missed pass delays delivery to the people who need the data.
- Science and research missions
- Research institutions and mission teams flying experiments, testing new technologies, and collecting scientific measurements. Reliable contact planning helps teams make the most of limited communications windows and bring valuable results back to Earth.
- Government and public-sector missions
- Agencies and operators supporting civil space programs, public infrastructure, environmental monitoring, and emergency response. These missions need confidence in communications availability, with clear visibility into coverage gaps and operational constraints.
- University, CubeSat, and small satellite teams
- Teams working with tight power budgets, limited ground access, and small operations crews—who need to understand whether a link will close, what a pass can deliver, and when to be ready.
- Ground stations and ground station networks
- Commercial providers, government facilities, and research stations supporting one spacecraft or many. Understanding visibility, link performance, and competing contact windows helps operators plan antenna time and deliver dependable service.
What is in the first release
Two model families are live. Our SNR models are trained on observations from ground- and space-based sensors around the world, including weather, atmospheric conditions, link margin, and measured SNR, and they forecast signal strength one to five minutes ahead. Our demand models learn from the traffic history of each pool to forecast incoming demand for the next hour, in ten-minute steps.
This release opens the Bronze tier of both model families (Silver and Gold models remain locked to Enterprise tiers).
You can call both from your own software with a scoped API key, and the quick start takes you from a new key to a first forecast in about five minutes. Claude Code, Cursor, Codex, and other MCP clients can read your topology, request forecasts, and send telemetry, and the Agents page writes the config for each one with your key filled in. Integrations send every forecast to Grafana Cloud as metrics or to your own Amazon S3 bucket, at no cost to your plan.
To try it without your own data, use the demo fleet of 6 ground stations, 39 satellites, and 21 links. You can load it into your account through an agent, or download it as an open dataset on Hugging Face.
For students and researchers
Academic access ships with this first release. Sign up or link a school email, and your account gets Developer allowances for free, no card required.
Start at Constellation for Education, or email education@constellation.space to bring OS Platform intelligence to your program.
Where the platform is going
Forecasting is the first part of a larger system, a control plane for fleets that span many operators, networks, and orbits. We are building it in three stages.
Available now
Monitor and predict
Ingest fleet telemetry and forecast link quality and traffic. This is what the first release delivers.
Next
Plan and optimize
Turn forecasts into decisions about which contacts to book, which ground station to use, and how to allocate capacity before demand arrives.
After that
Execute
Apply those decisions through integrations with customer systems and modems, and record every action so you can see what changed and why.
Enterprise customers already work with us across all three stages. As each stage matures, we will bring it to the Platform behind the same API and agent tools.
What we are building next
In training
New forecasting models
Models for conjunction screening, weather impact on links and passes, and contact planning, served through the same API and agent tools as the models in this release.
In development
Backtests and evaluations
Replay your recorded telemetry through any model and get a scorecard of its predictions against what actually happened, so you can see how it performs on your own fleet before it goes into operations.
In research
Space operations language models
Our research team is tuning language models for space operations.
Get started
Sign in to the Constellation Platform, watch the short tour, and make your first forecast.
To keep up with new product and research announcements, follow us on X and LinkedIn, and read the Constellation blog.