Kong Ai Gateway Bridge

naftiko: "1.0.0-alpha2"

info:
  title: Kong AI Gateway Bridge
  description: >-
    Routes Naftiko-side LLM calls through Kong AI Gateway's universal LLM
    API (OpenAI / Anthropic / Gemini / Bedrock / Azure / Databricks /
    Mistral / HuggingFace under one OpenAI-compatible surface) — automatic
    semantic caching, token budgets, prompt firewalls, PII guardrails,
    automated RAG injection, and per-token observability with no
    per-capability LLM-vendor wiring.
  tags:
    - Naftiko
    - Kong
    - Partnership
    - AI-Gateway
    - LLM-Proxy
    - Token-Budget
  created: '2026-05-15'
  modified: '2026-05-15'

binds:
  - namespace: kong-ai-env
    description: Kong AI Gateway endpoint + token.
    keys:
      KONG_AI_BASE: KONG_AI_BASE
      KONG_AI_TOKEN: KONG_AI_TOKEN

capability:
  consumes:
    - namespace: kong-ai
      type: http
      baseUri: '{{KONG_AI_BASE}}'
      authentication:
        type: bearer
        token: '{{KONG_AI_TOKEN}}'
      resources:
        - name: chat-completions
          path: '/v1/chat/completions'
          operations:
            - name: chat-completion
              method: POST
        - name: completions
          path: '/v1/completions'
          operations:
            - name: completion
              method: POST
        - name: embeddings
          path: '/v1/embeddings'
          operations:
            - name: embedding
              method: POST
        - name: list-models
          path: '/v1/models'
          operations:
            - name: list-models
              method: GET
        - name: token-budgets
          path: '/v1/ai/token-budgets'
          operations:
            - name: list-token-budgets
              method: GET
            - name: create-token-budget
              method: POST
        - name: token-usage
          path: '/v1/ai/usage'
          operations:
            - name: get-token-usage
              method: GET
              inputParameters:
                - { name: since, in: query, required: false }
                - { name: model, in: query, required: false }
                - { name: agent, in: query, required: false }

  exposes:
    - type: rest
      address: 0.0.0.0
      port: 8080
      namespace: kong-ai-gateway-bridge-rest
      description: OpenAI-compatible REST surface backed by Kong AI Gateway with semantic cache, token budget, prompt firewall layered in.
      resources:
        - name: chat-completion
          path: '/v1/chat/completions'
          operations:
            - name: chat-completion
              method: POST
              call: kong-ai.chat-completion
        - name: completion
          path: '/v1/completions'
          operations:
            - name: completion
              method: POST
              call: kong-ai.completion
        - name: embedding
          path: '/v1/embeddings'
          operations:
            - name: embedding
              method: POST
              call: kong-ai.embedding
        - name: list-models
          path: '/v1/models'
          operations:
            - name: list-models
              method: GET
              call: kong-ai.list-models
        - name: token-budgets
          path: '/v1/ai/token-budgets'
          operations:
            - name: list-token-budgets
              method: GET
              call: kong-ai.list-token-budgets
            - name: create-token-budget
              method: POST
              call: kong-ai.create-token-budget
        - name: token-usage
          path: '/v1/ai/usage'
          operations:
            - name: get-token-usage
              method: GET
              inputParameters:
                - { name: since, in: query, type: string, required: false }
                - { name: model, in: query, type: string, required: false }
                - { name: agent, in: query, type: string, required: false }
              call: kong-ai.get-token-usage

    - type: mcp
      address: 0.0.0.0
      port: 3010
      namespace: kong-ai-gateway-bridge-mcp
      description: MCP server exposing Kong AI Gateway's universal LLM proxy as agent-callable LLM tools.
      tools:
        - name: chat-completion
          description: Run an OpenAI-compatible chat completion through Kong AI Gateway (semantic cache + token budget + prompt firewall + PII guardrails).
          hints: { destructiveHint: false }
          call: kong-ai.chat-completion
        - name: completion
          description: Run an OpenAI-compatible text completion through Kong AI Gateway.
          hints: { destructiveHint: false }
          call: kong-ai.completion
        - name: embedding
          description: Compute embeddings through Kong AI Gateway.
          hints: { readOnly: true }
          call: kong-ai.embedding
        - name: list-models
          description: List the LLM models available through Kong AI Gateway (provider-agnostic).
          hints: { readOnly: true }
          call: kong-ai.list-models
        - name: list-token-budgets
          description: List Kong AI Gateway token budgets (per-agent / per-period token caps).
          hints: { readOnly: true }
          call: kong-ai.list-token-budgets
        - name: create-token-budget
          description: Create a new Kong AI Gateway token budget.
          hints: { destructiveHint: false }
          call: kong-ai.create-token-budget
        - name: get-token-usage
          description: Get Kong AI Gateway token consumption stats per model / agent / period.
          hints: { readOnly: true }
          inputParameters:
            - { name: since, type: string, required: false }
            - { name: model, type: string, required: false }
            - { name: agent, type: string, required: false }
          call: kong-ai.get-token-usage