Service level agreement
Last updated: 2026-06-23. Plain-language. The lawyer-formal version is at /legal/sla-formal.
The promise
| Surface | Monthly uptime target | Measurement window |
|---|---|---|
Inference (api.greenjoules.cloud/v1) | 99.5% | Calendar month |
| Object Store | 99.9% availability, 99.999999999% (11 nines) durability | Calendar month |
| JouleDB | 99.5% | Calendar month |
| Compute / Functions | 99.5% (single-region), 99.9% (multi-region region=auto) | Calendar month |
| Console / portal | 99.9% | Calendar month |
Note: v1 launches at the above targets. We are targeting 99.9% for Inference + JouleDB by Q1 2027 once the mesh reaches 50 nodes.
How we measure
From multiple independent vantage points (Cloudflare workers in 8 cities + an on-platform synthetic) probing the public endpoint every 60 seconds. A request counts as "down" if it fails or times out for two consecutive probe cycles within a region; otherwise it's up. Probe results are continuously published at status.greenjoules.cloud.
What doesn't count as downtime
- Scheduled maintenance (announced >= 7 days in advance, planned during your account's low-traffic window).
- Disruption from your action: an exhausted balance, a workload that hit its
energy_budget, a bug in your own code. - Force-majeure events: large-scale internet routing outages outside our control, sanctioned governmental orders.
- Third-party model outages: if you're pinned to
claude-haiku-4.5and Anthropic's API is down, that doesn't count against the Inference SLA. Use"auto"to fail over.
What you get if we miss
SLA credits applied automatically as a balance top-up the following month:
| Monthly uptime | Credit |
|---|---|
| < 99.5% but >= 99.0% | 10% of monthly bill, that surface |
| < 99.0% but >= 95.0% | 30% of monthly bill, that surface |
| < 95.0% | 50% of monthly bill, that surface |
Credits cap at 100% of the affected month's bill for that surface. Credits don't expire and apply to any future workload.
How to claim
You don't. If our monitoring detects the breach, the credit lands automatically. If you think we missed something, write to [email protected] within 30 days of the affected month's end with the workload IDs and time windows; we'll investigate and credit if confirmed.
Incident communication
- Public status: status.greenjoules.cloud, updated within 5 minutes of an automated alert.
- Customer email: if an incident affects your account directly, you get an email within 15 minutes of detection.
- Postmortems: for incidents that affect >1% of customers or last >30 minutes, a public postmortem within 7 days at
/postmortems/<incident-id>.
Enterprise SLA
For workloads needing > 99.9% target, contractual penalties beyond credits, or 24/7 named-engineer response, contact [email protected] for the enterprise SLA tier.