SLA · v0.2 alpha
Service-level agreement.
Three tiers, three honest commitments. Read this twice. Single-site and single-operator constraints are disclosed up front and apply to every tier until Q4 2026 multi-site rollout.
Per-tier commitments.
Open Shard · per-token
Best-effort. No 99.x uptime commitment.
Open Shard exists to qualify buyers and serve non-critical workloads. We do not commit to an uptime number on this tier and we will not refund Open Shard credit for downtime — that is the trade for the price.
- Uptime target
- Best-effort, typically >98% measured monthly. Not contractual.
- Response time
- None committed. Status questions answered on best-effort basis.
- Refund policy
- None. Use this tier when you can tolerate occasional 503s.
- Failover
- Within the fleet only. No multi-site failover until Q4 2026.
Dedicated Shard · per-hour
Best-effort 99.0% during reserved window.
You reserve a specific Blackwell node for a specific window. We do not share the card with anyone else while you hold it. If our power or upstream goes down inside your window, we credit you for the lost time.
- Uptime target
- 99.0% during your reserved hours, measured per-window.
- Response time
- Operator pages within 1 hour for any incident lasting >5 minutes.
- Refund policy
- Pro-rated credit for any downtime exceeding 1% of the reserved window. Credits issued automatically; visible on next invoice.
- Failover
- Cross-node within tampa-1 where workload allows. Disclosed at reservation time if not.
- Pager
- One human operator on call. iJarvis founder/CTO.
Private Fleet · monthly contract
Negotiated. Default template below.
Private Fleet SLA is written into the contract. The default template — used unless we negotiate something stronger or weaker — looks like this:
- Uptime target
- 99.5% during contracted hours, with credits at 99.0% and 95.0% breach thresholds.
- First response
- 4 hours, 24/7. Named operator slack channel.
- Refund policy
- Tiered credits per breach threshold; capped at 30% of monthly fee in any single month.
- Data residency
- US-only, written into contract. No cross-border traffic without explicit amendment.
- Quarterly review
- Operator + customer review of throughput, incident history, capacity planning.
What no SLA covers.
- Single-site failure modes until Q4 2026 multi-site split. If our Tampa site loses power and our backup capacity is exhausted, all tiers go dark.
- Third-party outages we route through (Tailscale control plane, Cloudflare DNS / Pages, upstream ISP, NVIDIA driver bugs). We will communicate, but the SLA clock pauses for documented third-party incidents.
- Customer-side issues — model checkpoint corruption, your code making malformed requests, your prompts hitting your own rate-limit ceilings.
- Force majeure in the conventional sense.
How to escalate.
- Open Shard: email
hello@ijarvis.aiwith subjectiShard Open Shard incident. Best-effort response. - Dedicated Shard: Slack channel issued at reservation time. The operator is the only on-call human.
- Private Fleet: Slack channel named in your contract. Phone number provisioned at signing for tier-1 incidents.
Change log.
v0.2.0— 2026-05-07 — first published version. Three-tier framework. Single-site disclosure. To be revisited at multi-site rollout.
Honest about v0.2
These terms govern the alpha period. Material changes will be sent to active customers 30 days before they take effect. The cousin who reads everything looking for liars helped harden this page; the gaps that remain are gaps because we have not earned the claim yet, not because we forgot to write it down.