NeuralTrust | Platform for Agent Security.

Clients send the same JSON they would send to Cohere /v2/rerank. Auth and the consumer slug are the same as chat.

curl -X POST "https://<gateway-host>/<consumer-slug>/v1/rerank" \
  -H "X-AG-API-Key: <your-api-key>" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "rerank-english-v3.0",
    "query": "What is TrustGate?",
    "documents": ["TrustGate is an LLM gateway", "Unrelated text"]
  }'

On a Private data plane, add X-AG-Gateway-Slug as in the chat snippets. The gateway speaks the Cohere v2 rerank shape on this path. It is not an OpenAI Chat Completions route.

Providers

Registry	Upstream
Cohere	Cohere `/v2/rerank`

OpenAI, Azure, Anthropic, Gemini / Vertex, Bedrock, Groq, Mistral, OpenRouter, and other chat-only providers do not expose rerank. They are left out of the candidate pool. The model field is allowlisted as usual and can pin with @provider/model.

Routing

The consumer’s registries are filtered to those that advertise rerank.
Model resolution and load balancing run on that pool.
Pinning a registry that cannot rerank is a 400, not a failover.
An empty capable pool is a 503.

Attach at least one Cohere registry to the consumer. A mixed chat + rerank pool is fine: chat still uses every chat-capable member; rerank skips the rest.