WITHOUT QORTEX
HIGH BY DEFAULTOne model for everything.
01Routine request02Coding request03Complex request
EVERY REQUESTYour strongest modelHighest cost every time
AI ROUTING. BRING YOUR OWN KEYS.
YOUR MODELS.
ONE API.
LESS WASTE.
Not every kind of work needs the same model. Define policies for coding, research, extraction, or your own workloads. Qortex chooses the policy and effort; you control every model.
You pay providers for inference. Qortex manages the calls.
REQUEST.ROUTEILLUSTRATION
01 / YOUR APPLICATION
“Extract the dates from this text.”
One Qortex API key↓
02 / YOUR BYOK ROUTER
A straightforward extraction task.
Choose Data extraction / Low.
↓
LowMediumHighMaxUltra
03 / SELECTED MODEL
Your lower-cost model↗
Your lower-cost model↗
Router + answering model = 2 outbound calls
01 YOUR PROVIDER KEYS02 YOUR MODEL CHOICES03 ONE APPLICATION KEY04 CLEAR CAPACITY PRICING
THE PROBLEM QORTEX SOLVES
STOP USING YOUR MOST EXPENSIVE MODEL FOR EVERY REQUEST.
Routine extraction and support questions rarely need the same intelligence as difficult coding or research. Using one powerful model for everything makes simple work unnecessarily expensive.
WITH QORTEX
RIGHT-SIZEDUse strength where it matters.
01Routine request→LOW EFFORTEfficient model
02Coding request→CUSTOM POLICYYour coding model
03Complex request→MAX EFFORTStrong model
04Model failure→FALLBACKYour backup model
Qortex routes each request through the workload policy and model tier you chose.Use cheaper models for routine work. Escalate only when your rules require it.See how it works
A SHORTER PATH TO MODEL ROUTING
THREE STEPS.
ONE CONNECTION.
Keep your application. Put Qortex between it and your AI providers.
Bring your keys.
Add your existing provider API keys. Model usage is still billed to those provider accounts.
Provider connectionsYour keys →
Define workload policies.
Create policies such as Coding or Research, then choose a primary and backup model for every effort level.
Policy + effortYour models →
Use one Qortex key.
Let the router choose policy and effort, provide either one explicitly, or call a provider model directly.
Your applicationOne key →
AUTOMATIC DOES NOT MEAN OUT OF YOUR HANDS
YOU SET THE RULES.
QORTEX HANDLES THE CALLS.
POLICY
LowYour fast coding model
MediumYour everyday coding model
HighYour stronger coding model
MaxYour complex-work model
UltraYour strongest setup
Different work.
Different model ladders.
Create a policy for each kind of work, with its own model ladder. The router chooses the policy and effort; you remain in control of every model.
FALLBACK.CHAINEXAMPLE
Primary modelUnavailable
↓
Tier backupTry next
↓ if exhausted
Global last resortFinal attempt
A backup plan.
Not a dead end.
Set fallbacks per tier, then a global last resort. Apply attempt limits and restrictions. If no eligible model works, get a clear failure.
SESSION.ROUTEEXAMPLE
First step Router → Model A
Tool result→ Model A
Next step→ Model A
Next step→ Model A
KEEP THE MODEL. REUSE THE PREFIX.
Long runs.
Less routing overhead.
Sticky sessions skip repeated router calls and keep the same provider and model, helping later steps reuse provider prompt caching when supported. Cache eligibility and savings depend on the provider.
CAPACITY, WITHOUT THE GUESSWORK
Busy? Queue it.
Ready? Run it.
Limit simultaneous calls and schedule waiting requests around provider limits. More Qortex capacity does not increase your provider quota.
EXECUTION.QUEUEILLUSTRATION
Running01020304
Waiting05060708
Your concurrency limit controls what runs next.
PAY FOR CAPACITY. NOT ANOTHER TOKEN MARKUP.
START AT $10.
GROW AS YOU GO.
One workspace subscription. Your provider bills stay separate.
Qortex billRouting + execution capacityProvider billsAll model inference, including the router
Starter includes 250,000 outbound provider calls. Growth includes 1,000,000. Increase capacity only when you choose.
STARTERMONTHLY / USD
$10/ month
Includes 250,000 outbound provider calls.
- BYOK policy and effort routing
- Per-policy model ladders and fallbacks
- Queues and concurrency controls
- Standard bandwidth and 30-day operational metadata
$10 / month
Growth is $25 for 1,000,000 calls. Each additional 1,000,000 calls adds $20. Extra capacity is opt-in.
Understand how calls are counted ↓NO SURPRISES IN THE COUNTER
ONE REQUEST CAN
MAKE MORE THAN ONE CALL.
Each provider attempt Qortex dispatches counts: routing, answering, review, retry, or fallback. Cached provider requests count too, even when their inference cost is lower.
Automatic selection1 router + 1 answer = 2 calls
Sticky session continuation1 answer = 1 call
Tools run in your application0 Qortex model calls
A retry or fallback counts only when Qortex dispatches it to a provider. Requests rejected before dispatch and requests waiting in the queue do not consume call allowance.
YOUR DATA. YOUR CHOICE.
METRICS BY DEFAULT.
CONTENT BY PERMISSION.
Track costs, timing, and errors without keeping prompt history. Private routing optimization would be a separate opt-in feature, with retention and evaluation budgets you control.
Durable jobs temporarily retain the payload needed to run them. Provider retention policies still apply. Shared-model training requires separate authorization.
A FEW THINGS TO KNOW
GOOD QUESTIONS.
CLEAR ANSWERS.
Does $10 include model tokens?
No. It pays for Qortex's software and included execution capacity. Every AI model call—including routing and review—is billed through your connected provider keys.
Do I have to route every step?
No. Specify a policy and effort directly, provide only one for Qortex to choose the other, or use a sticky session to reuse the selected route across a long run.
Will provider prompt caching still work?
Qortex is designed to preserve supported provider caching fields and prompt prefixes. Cache eligibility, lifetime, and pricing belong to the provider. Changing models or waiting too long may lose cache reuse.
Does Qortex execute my tools?
No. Qortex returns the model response, including supported tool-call requests. Your application executes the tools and sends their results back.
Is a 90% cost reduction guaranteed?
No. Savings depend on the workload, model pool, router overhead, caching, and quality requirements. There is no published Qortex benchmark yet; percentage claims will require reproducible measurements.
Can I use Qortex without other products?
Yes. Qortex is a standalone product. It does not require osFoundry, osModels, or osAgents.