RouteOne hostname. Two provider-native surfaces.
The intelligent model gateway · by Zimac.AI
Every model has a door. Sage holds the keys.
Point your existing SDKs at Sage. She keeps provider credentials off every laptop, checks identity, entitlement, and spend before the call, then meters the route and can fail over before the first byte.
Anthropic + OpenAI protocols · hosted or entirely in your cloud
ProtectProvider credentials never reach a laptop.
GovernIdentity, entitlement, and spend checked first.
AccountUsage lands on a seat, org, and model.
Sage · Model Gateway
The judgment at the threshold.
Sage personifies the control plane around every model call: who is asking, which models they may use, what the call can spend, where it should run, and how its usage is accounted for.
“The right model. The right wire. The right limit.”
- Know the callerA revocable seat token resolves to a person and org.
- Honor the wireAnthropic stays Anthropic. OpenAI stays OpenAI.
- Own the spendUsage stays attached to the seat, org, and model served.
One name · both native wires
Change the address. Keep the protocol.
Change the base URL and replace the provider key with a per-user Sage token. Your Anthropic client still speaks Anthropic; your OpenAI client still speaks OpenAI. Identity, spend checks, metering, and routing surround both.
client = Anthropic(
base_url="https://sage.zimac.ai",
api_key="zmc-your-seat-token"
)
client.messages.create(
model="claude-sonnet-4-6", ...
)client = OpenAI(
base_url="https://sage.zimac.ai/v1",
api_key="zmc-your-seat-token"
)
client.chat.completions.create(
model="gpt-5", stream=True, ...
)Control before compute
She lets the right call through.
The request is the easy part. The policy, money, routing, and evidence around it are the product.
One seat. One revocable token.
Every caller resolves to a person and an org. Provider secrets stay server-side, while seat tokens can be minted, edited, and revoked without rotating provider keys across the fleet.
zmc••••••••••7FD2FINANCE · ACTIVE
Spend checked before compute.
Hosted calls reserve a conservative upper bound from prepaid credit before reaching the provider, then settle against reported usage. Seat and org limits are completed-spend caps; prepaid credit is the hard cross-replica floor.
$118.42 remaining
Fail over before the first byte.
Pin each model ID to its own upstream and optional fallback. Sage switches on eligible upstream failures only before response bytes begin.
PRIMARYFALLBACK
Usage lands on an owner.
Provider-reported usage is read as bytes stream past and priced from the effective rate table—per org, seat, and model, ready for showback or chargeback.
12.4Mtokens$84.19month
Only the models this org may use.
The entitlements that gate each request also shape /v1/models, so clients see only the catalog available to their organization.
sonnetgpt-5embed
Protocol-true by design
The gateway should not become the risk.
Sage adds the enterprise control plane without flattening every provider into a lowest-common-denominator API. The native surfaces remain recognizable, and each amendment is narrow and explicit.
Read the technical foundations →No universal translation layer
Anthropic uses the Messages surface. OpenAI uses Chat Completions and compatible endpoints.
One static Rust binary
Distroless, non-root, with streaming-safe rollouts, metrics, probes, and an admin console.
Money is a first-class object
Prepaid reservation, usage settlement, caps, and metrics stay on one accountable path.
Two ways through
Hosted in minutes. Or entirely yours.
SAGE · HOSTED
Pay as you go
Create an account, invite the org, mint seat tokens, and top up one shared prepaid balance. The published rate table is the meter.
- Provider credentials stay server-side
- No per-seat gateway fee
- Shared credit with seat and org caps
SAGE · SELF-HOSTED
Your cloud
Run the same gateway in your AWS account alongside Zimac Enterprise. Your provider keys, traffic, and metrics stay yours.
- Flat annual enterprise plan
- Unlimited seats
- Helm + Terraform deployment
Sage is the gateway
Give every seat a token.
Keep every key at home.
One governed threshold for every model your organization uses.