API
Rerank and moderation
Use transient scoring and safety inference through the same boundary
Rerank and Moderations are separate route-scoped inference operations. They share workspace policy and provider isolation but neither creates a search index, document database or safety case-management system.
Rerank candidates
POST /v1/rerank scores transient candidate strings or objects against a query. The documents protocol field is not persisted or connected to an Octoryn Documents or Knowledge layer.
{
"model": "provider/rerank-model",
"query": "Which control prevents cross-tenant access?",
"documents": ["Workspace-bound identity", "Client-side model list"]
}Rerank usage and fallback
Eligible provider preferences and fallback apply to rerank-capable endpoints. Usage may include token and request/search-unit facts; consumer search billing is intentionally outside Router.
Moderate input
POST /v1/moderations forwards a compatible moderation request to an eligible configured provider and normalizes the response boundary. The Router does not invent a provider-independent legal or product policy verdict.
Treat results as application inputs
Applications remain responsible for their policy thresholds, user handling and human review. Do not log sensitive candidate or moderation content merely to explain routing.