Operations
During an incident
What to do, and what to expect from Volt, during a service incident.
This page is for customers operating on Volt during an incident.
Check status first
- Status page: voltcloud.ai/status — current state per service and SKU.
- Incidents are posted with a severity, the affected metros/SKUs, and updates on a cadence.
What Volt guarantees
- 99.9% uptime SLA (Tier III standard).
- Service credits at breach thresholds: 99.0% and 98.0% monthly availability.
- During a sovereign-tier incident, your data still never leaves its metro — zero egress holds even in failure.
What your app should do
The SDK is safe by default; lean on it:
- Bounded retries are automatic — 429 honors
Retry-After, 503 backs off. Override withmax_retries=. - Streams resume on disconnect when the pod's KV cache is warm, otherwise re-issue.
- Catch typed errors and read
request_id/pod_idfor support:
from volt import Volt, NoCapacity, VoltError
try:
resp = client.chat.completions.create(model="llama-3.3-70b-instruct", messages=[...])
except NoCapacity as e:
# No eligible pod right now — back off and retry, or fail over.
log.warning("no capacity", request_id=e.request_id)
except VoltError as e:
log.error("volt error", err=str(e), pod_id=e.pod_id)Sovereign and pinned workloads
If you pin a metro, a capacity event in that metro will surface as NoCapacity
rather than silently serving elsewhere — that is by design. Decide in advance
whether to wait or to relax the pin for non-sensitive traffic.
Opening a support ticket
Include the request_id and pod_id from the error, the model, the metro, and the
timestamp. Customers on a support SLA reach support per their contract; otherwise
email support@voltcloud.ai. Security issues go to security@voltcloud.ai — never
a public channel.