One schema across 13 providers
OpenAI-compatible REST and streaming. Completions, tool-calling, vision, and structured JSON ride one code path; provider dialects are normalized at the gateway, never in your adapter code.
FIG.01 — UNIFIED INFERENCE GATEWAY
One endpoint.
Every frontier
model.
One API schema across 36 models from 13 providers, routed to the healthiest replica under a written SLA.
SEC.02 — ROUTING
Thirteen upstreams. One contract.
Requests enter as one schema. The gateway scores replicas, dispatches to the healthiest pool, and reports the route back in headers.
One schema in. Provider dialects end at the gateway.
Unhealthy pools drop out before your retry budget does.
Route, rate, and latency come back in X-IDC-* headers.
SEC.03 — SPECIFICATIONS
What the gateway guarantees.
SLA-aware routing & failover
Requests land on the lowest-latency healthy replica in your region. When an upstream saturates, traffic shifts to the next pool before your retry budget runs out.
Pricing that survives finance review
Per-token rates are quoted against your workload, then locked for the term once you commit — along with the dedicated capacity behind them. No per-region multipliers hiding in a PDF.
Response-header observability
Every response carries its route, provider latency, and the rate you paid in X-IDC-* headers. Stream traces to your own pipeline; there is no dashboard you must live in.
SEC.04 — INDEX OF MODELS
Eight of thirty-six.
A cross-section of the catalog, updated the day upstream ships. Rate bands are indicative, not prices.
| Model | API ID | CTX | Rate |
|---|---|---|---|
| Claude Fable 5 | claude-fable-5 | 1M | $$$$ |
| Claude Opus 5 | claude-opus-5 | 1M | $$$ |
| GPT-5.6 Sol | gpt-5.6-sol | 1M | $$$ |
| Gemini 3.6 Flash | gemini-3.6-flash | 1M | $$ |
| Grok 4.5 | grok-4.5 | 500K | $$ |
| DeepSeek V4 Pro | deepseek-v4-pro | 1M | $ |
| Kimi K3 | kimi-k3 | 1M | $$$ |
| Llama 4 Maverick | llama-4-maverick | 1M | $ |
$ entry · $$ standard · $$$ premium · $$$$ frontier · band, not a price · quoted per workload
View full catalogSEC.05 — FAQ
Asked before signing.
Q.01What is idclinktech?
An inference gateway: one OpenAI-compatible endpoint in front of 36 models from 13 providers, with routing, failover, and billing handled below your code. We do not train models and do not want to. We keep the ones you depend on reachable.
Q.02How does routing pick a model host?
Closed-weight IDs resolve to their provider; open-weight IDs resolve to managed pools. Within that, the gateway scores replicas on rolling error rate and p99 latency in your region and dispatches to the healthiest one. The chosen route is reported in the X-IDC-Route response header.
Q.03Can I pin a provider?
Closed models are pinned by definition: claude-opus-5 is served by Anthropic, nowhere else. For open-weight models served from managed pools, Scale and Enterprise plans can lock a route per model ID; the pin is enforced at dispatch and visible in response headers.
Q.04What happens when a provider retires a model ID?
The ID leaves the catalog and enters an alias window: the gateway keeps resolving it to its documented successor for a published period, the way a 301 works for URLs. Every removal, alias window, and successor ID is recorded in the changelog before it takes effect.
Q.05How do I get an API key?
Email sales@idclinktech.com with your workload shape: expected volume, the models you care about, and any compliance constraints. You will typically have a trial key and a pricing sheet for your shape within one business day.
SEC.06 — COMMISSIONING
Boring infrastructure
for an exciting industry.
The catalog changes monthly. Your integration does not have to.