Filtration Gateway

The Filtration Gateway is the security control plane. While mostly used internally by the Inference Gateway, it exposes endpoints for context resolution and prompt processing.

Context Resolution

POST /internal/context/resolve

Resolves an API key and model name into a fully configured inference context (deployment details, guardrails, RAG config).

api_key

The API key used for the request.

model

The requested model alias (e.g., gpt-4).

Response:

{
  "valid": true,
  "deployment": {
    "id": "uuid",
    "endpoint": "http://worker:8000"
  },
  "guardrail_config": { ... },
  "rag_config": { ... }
}

Prompt Processing

[!NOTE] These endpoints are primarily for internal service communication.

POST /internal/prompt/process

Applies prompt templates and retrieves RAG context before inference.

messages

Original list of messages.

template_id

(Optional) Specific template ID to apply.

rag_config

Configuration for RAG retrieval during this turn.

Response: Returns a modified list of messages including system prompts and injected context.