Concepts
Sovereignty
In-metro inference, the sovereign tier, and client-side enforcement.
Sovereignty on Volt means your inference runs in a specific metro, your data never leaves it, and you can prove it. It is enforced at three layers: the data plane (zero egress), the control plane (sovereignty profile), and the SDK (client-side).
The sovereign tier
Spark offers two tiers:
| Tier | Llama 70B price | What you get |
|---|---|---|
| Standard | $0.95/M tokens | In-metro serving, zero egress |
| Sovereign | $1.45/M tokens | Pod-pinned inference, dedicated SPIRE trust domain, extended audit retention, compliance attestation pack |
The sovereign tier is ~55% above standard — still 45% below AWS Bedrock ($2.65/M).
Turning it on
Set sovereign=True and pin a metro. The SDK then enforces it as a defense layer:
from volt import Volt, SovereigntyViolation
client = Volt(sovereign=True, pinned_metro="us-east-iad")
try:
resp = client.chat.completions.create(
model="llama-3.3-70b-instruct",
messages=[{"role": "user", "content": "Summarize this contract."}],
)
print(resp.volt.tier, resp.volt.metro) # verified before you see the data
except SovereigntyViolation:
... # response payload is withheld from your codeThree layers of enforcement
- Data plane — zero egress makes residency physical.
- Control plane — your org's sovereignty profile pins tier, metro, and allowed catalogs.
- SDK — every response is validated client-side; a mismatch raises and withholds the payload.
Attestation
Vault customers fetch a measured-boot attestation report per rack before deploying sensitive workloads:
voltctl attest --rack rack-iad-7