Routing plane
Classify sensitivity and task, apply policy, then select the best eligible model across every dimension that matters, including cost, latency, quality, and fit, and dispatch synchronously on every request.
Architecture
Inside your perimeter
Cardix runs as a single service within your network. Applications connect via an OpenAI-compatible endpoint; Cardix handles classification, policy, selection, and dispatch.
Multi-axis
Optimization axes
Cost, latency, quality, compliance, personalization, and custom metrics you define
Hybrid
Deployment modes
On-prem, VPC, and air-gapped
1
OpenAI-compatible endpoint
Drop-in for your existing SDK
Fail-closed
Policy enforcement
Residency rules before routing
Planes
Three layers, one purpose
Routing, learning, and observability work together without exposing your data or blocking your responses.
Learning plane
Captures traffic signals and the custom metrics you define, then builds per-team and per-workload model-fit intelligence to improve selection over time. Runs asynchronously and never blocks the response.
Observability plane
Full audit trail, spend vs baseline comparison, and visibility into blocked demand.
Request journey
What happens on every request
Five steps, milliseconds. Policy before routing, every time.
- 01
Authenticate
Validate the caller and attribute the request to your organization.
- 02
Classify
Determine data sensitivity and task category for the prompt.
- 03
Policy
Evaluate residency rules and produce allowed deployment locations. Fail closed if no rule matches.
- 04
Select
Weigh every dimension you care about, including cost, latency, quality, personalization, and custom metrics, across your eligible catalog, then choose the best model for this request.
- 05
Dispatch
Route to the chosen backend and return the response, streaming or buffered.
Learning & personalization
Routing that gets better with every request
Cardix does not stop at a static rule. It learns from your traffic and the custom metrics you define, then continuously improves model selection for your organization, all inside your perimeter and off the response path.
01
Observe
Every routing decision records its signals: the chosen model, cost and latency actuals, quality outcomes, and the custom metrics your organization defines.
02
Learn
Cardix builds model-fit intelligence for your organization, learning which models perform best for each team, task type, and workload, entirely on your infrastructure.
03
Optimize
That intelligence refines how Cardix weighs each axis, so selection reflects what actually works for your traffic rather than a static, one-size-fits-all default.
04
Improve
The next matching query routes better. The loop runs continuously and asynchronously, off the hot path, so it never adds latency to a response.
Continuous, asynchronous, and off the hot path. The loop never adds latency to a response.
Signals captured
Task & sensitivity class
The classification assigned to each request, so learning is grouped by the work being done.
Per-team & per-workload outcomes
How models perform for different teams and workloads across your organization.
Cost & latency actuals
The real cost and speed of each decision, measured on your infrastructure.
Quality outcomes
Automated quality signals and any application or human feedback you choose to send.
Custom organization metrics
Your own quality bars or business KPIs. You define what matters, and Cardix optimizes for it.
All signals stay inside your perimeter. The intelligence is built on your infrastructure. Your data never leaves your network, and nothing is sent back to us.
Integration
Three steps to intelligent routing
No rip-and-replace. Cardix integrates with the stack you already run.
01
Point your SDK
Change the base URL in your existing OpenAI-compatible client. No application rewrite, and streaming is supported.
02
Set your policy
Define residency rules, sensitivity labels, and approved backends. Cardix enforces them before any model is considered.
03
Route across your fleet
Every request is classified, optimized, and dispatched to the best eligible model across on-prem, private cloud, or approved APIs.
Get started
Deploy intelligent routing inside your perimeter
Review the architecture, assess fit for your stack, and speak with our team about your deployment requirements.