Model-agnostic infrastructure
Yes—you can use your own models and private endpoints. The model is a component, not your architecture.
Models change. Your system shouldn't have to.
Who model neutrality is for
Teams that expect the model to change: a better release next quarter, a cheaper model for classification, a private endpoint for data that cannot leave the network. If the agent is written against one vendor SDK, every switch becomes a migration.
How routing works
Assign an endpoint to an agent, or send the same task through more than one endpoint. Obliq records each run. You compare the traces—quality, latency, cost, and groundedness—and keep the agent definition. Sample comparison charts on the homepage are illustrative of the interface, not published benchmarks.
Same task, multiple models
SAME TASK
│
┌────────────┼────────────┐
↓ ↓ ↓
MODEL A MODEL B MODEL C
│ │ │
└────────────┼────────────┘
↓
COMPAREComparison metrics in product demos are labeled illustrative unless backed by published measurements. Use your own executions when you need a number you can stand behind.
Questions
Can I use my own models or private endpoints?
Yes. Obliq is architecturally model-agnostic. You can connect leading frontier APIs with your own keys, or private and open-weight endpoints on infrastructure you run, through a common runtime.
What is model routing?
Routing sends the same task to more than one model endpoint and keeps the agent definition fixed. You compare quality, latency, cost, and groundedness on the execution, then keep the endpoint that fits the job.
Will switching providers force an agent rewrite?
It should not. Instructions, knowledge scope, and tool policy live on the agent. The model is an assigned component. Comparison traces show what changed when the endpoint changed.