How GreenPT decides when your agent runs
Every scheduled run asks one question first: is now a clean moment on the grid where this model runs? If a cleaner hour is coming soon enough, the run moves there. Otherwise it runs now.
On every run
- 1 · Agent
Run starts
A Foundry schedule starts your agent. Its first action is the carbon gate.
- 2 · GreenPT
Find the grid
From the agent’s model deployment: exact, in zone, or an estimate.
- 3 · Forecast
Look ahead
Compare now with every slot up to the agent’s longest wait.
- 4 · Decision
Run now or move
Move only if it saves at least the minimum. The moved run starts once, then is cleaned up.
Three grid labels, always shown
Regional deployments
Standard or provisioned in one region: Azure keeps the model in that country, so the grid is known.
Data Zone deployments
Guaranteed inside the EU or the US. Our hourly probes show where inside the zone it runs.
Global deployments, unknown models
Azure may run the model anywhere. We use your resource’s region and say so.
Where Data Zone runs · our measurements
- Any EU region, Data Zone EU
- Nordics (Sweden)
- East US, East US 2, South Central US
- US East (Virginia)
- Central US
- US Central (Iowa)
- West US, West US 3
- US West
Microsoft guarantees the zone. Where inside it a model runs is our measurement, repeated every hour, so it is labelled as one.
What we promise
A run is never blocked
No data, no access, an error: the agent runs now.
A moved run happens once
Signed and single-use, so a replay or a copy is just checked again.
No API keys
Agents sign in with their own Microsoft Entra identity.
Decisions, not prompts
We store what we decided and why, never your task content.