Developers

Drop-in gateway contract

Qubanta is integrated as a proxy endpoint. Callers submit an inference request with optional constraints and policy hints. Qubanta returns a governed response: output plus stability state, confidence semantics, flags, and an action directive.

Architecture Reliability model

Proxy

Integration modes

Gateway mode: App → Qubanta → Provider. Observe-only (shadow evaluation) can be added later without changing control behavior.

Schema

Constraints as contracts

Structured outputs can be governed via JSON schema and required fields. Constraint failures collapse confidence deterministically and emit action=blocked.

Adapters

Provider independence

Provider adapters are thin. Qubanta does not require fine-tuning or proprietary hooks. The control surface remains stable across model churn.

Request / Response

POST /v1/infer
{
  "input": "...",
  "model": { "provider": "openai", "name": "gpt-4.1" },
  "context": { "task_type": "extraction" },
  "constraints": {
    "schema": { "...": "..." },
    "required_fields": ["id", "status"]
  },
  "policy": { "min_confidence": 0.45, "on_low_confidence": "escalate" }
}

{
  "output": "...",
  "confidence": 0.61,
  "reliability": "moderate",
  "state": "stable",
  "flags": [],
  "action": "accept"
}