# yaml-language-server: $schema=https://schema.zeabur.app/template.json
apiVersion: zeabur.com/v1
kind: Template
metadata:
    name: TensorZero Gateway
spec:
    description: TensorZero Gateway - AI inference optimization platform with model routing and observability
    icon: https://www.tensorzero.com/favicon.svg
    variables:
        - key: PUBLIC_DOMAIN
          type: DOMAIN
          name: Public Domain
          description: Your public domain for the gateway
        - key: OPENAI_API_KEY
          type: STRING
          name: OpenAI API Key
          description: Your OpenAI API key for accessing GPT models (required for default configuration)
        - key: ANTHROPIC_API_KEY
          type: STRING
          name: Anthropic API Key
          description: Your Anthropic API key for accessing Claude models (optional)
        - key: DEEPSEEK_API_KEY
          type: STRING
          name: DeepSeek API Key
          description: Your DeepSeek API key (optional)
        - key: GOOGLE_AI_STUDIO_API_KEY
          type: STRING
          name: Google AI Studio API Key
          description: Your Google AI Studio API key (optional)
        - key: MISTRAL_API_KEY
          type: STRING
          name: Mistral API Key
          description: Your Mistral API key (optional)
        - key: TOGETHER_API_KEY
          type: STRING
          name: Together API Key
          description: Your Together API key (optional)
        - key: VLLM_API_KEY
          type: STRING
          name: vLLM API Key
          description: Your vLLM API key (optional)
        - key: XAI_API_KEY
          type: STRING
          name: xAI API Key
          description: Your xAI API key (optional)
    tags:
        - AI
        - Gateway
        - Machine Learning
        - API
    readme: |
        # TensorZero Gateway

        TensorZero Gateway is an AI inference optimization platform that provides model routing, observability, and experimentation capabilities for AI applications.

        ## Features

        - **Model Routing**: Intelligently route requests to different AI models
        - **Observability**: Built-in metrics and monitoring with ClickHouse integration
        - **Experimentation**: A/B testing and variant management for AI models
        - **Multi-Provider Support**: Works with OpenAI, Anthropic, AWS Bedrock, and many other providers

        ## Configuration

        The gateway uses a TOML configuration file that defines:
        - Model configurations and routing rules
        - Function definitions for different AI tasks
        - Metrics collection and optimization settings

        ## Required Setup

        1. **API Keys**: Configure API keys for the AI providers you want to use
        2. **Configuration Files**: The service includes a sample configuration with GPT-4o-mini setup. Configure it in [Config Editor](https://zeabur.com/docs/en-US/data-management/config-edit).

        ## Health Check

        The service includes a built-in health check endpoint at `/status` that monitors the gateway's operational status.

        ## Documentation

        For more information, visit the [TensorZero documentation](https://www.tensorzero.com/docs/gateway/deployment/#configuration-1).
    services:
        - name: clickhouse
          icon: https://avatars.githubusercontent.com/u/54801242
          template: PREBUILT
          spec:
            source:
                image: clickhouse/clickhouse-server:24-alpine
            ports:
                - id: http
                  port: 8123
                  type: HTTP
                - id: native
                  port: 9000
                  type: TCP
            volumes:
                - id: data
                  dir: /var/lib/clickhouse
            instructions:
                - title: Clickhouse HTTP interface URL
                  content: ${ZEABUR_WEB_URL}
                - title: Clickhouse username
                  content: ${CLICKHOUSE_USER}
                - title: Clickhouse password
                  content: ${CLICKHOUSE_PASSWORD}
                - title: Clickhouse database
                  content: ${CLICKHOUSE_DB}
            env:
                CLICKHOUSE_DB:
                    default: zeabur
                    expose: true
                CLICKHOUSE_DEFAULT_ACCESS_MANAGEMENT:
                    default: "1"
                CLICKHOUSE_PASSWORD:
                    default: ${PASSWORD}
                    expose: true
                CLICKHOUSE_USER:
                    default: zeabur
                    expose: true
                PORT:
                    default: "8123"
                TENSORZERO_CLICKHOUSE_URL:
                    default: http://${CLICKHOUSE_USER}:${CLICKHOUSE_PASSWORD}@clickhouse.zeabur.internal:8123/${CLICKHOUSE_DB}
                    expose: true
            configs:
                - path: /etc/clickhouse-server/config.d/docker_related_config.xml
                  template: |-
                    <clickhouse>
                          <!-- Listen wildcard address to allow accepting connections from other containers and host network. -->
                        <listen_host>::</listen_host>
                        <listen_host>0.0.0.0</listen_host>
                        <listen_try>1</listen_try>

                        <logger>
                            <level>information</level>
                            <console>1</console>
                        </logger>
                    </clickhouse>
                  permission: null
                  envsubst: null
        - name: TensorZero Gateway
          icon: https://www.tensorzero.com/favicon.svg
          dependencies:
            - clickhouse
          template: PREBUILT_V2
          spec:
            source:
                image: tensorzero/gateway
                args:
                    - --config-file
                    - /app/config/tensorzero.toml
            ports:
                - id: web
                  port: 3000
                  type: HTTP
            configs:
                - path: /app/config/tensorzero.toml
                  template: |
                    # ┌────────────────────────────────────────────────────────────────────────────┐
                    # │                                   MODELS                                   │
                    # └────────────────────────────────────────────────────────────────────────────┘

                    [models."gpt-4o-mini-2024-07-18"]
                    routing = ["openai"]

                    [models."gpt-4o-mini-2024-07-18".providers.openai]
                    type = "openai"
                    model_name = "gpt-4o-mini-2024-07-18"

                    # ┌────────────────────────────────────────────────────────────────────────────┐
                    # │                                 FUNCTIONS                                  │
                    # └────────────────────────────────────────────────────────────────────────────┘

                    [functions.simple_llm_call]
                    type = "chat"

                    [functions.simple_llm_call.variants.baseline]
                    type = "chat_completion"
                    model = "gpt-4o-mini-2024-07-18"

                    # ┌────────────────────────────────────────────────────────────────────────────┐
                    # │                                  METRICS                                   │
                    # └────────────────────────────────────────────────────────────────────────────┘

                    [metrics.task_success]
                    type = "boolean"
                    optimize = "max"
                    level = "episode"
                  permission: 436
                  envsubst: true
            healthCheck:
                type: HTTP
                port: web
                http:
                    path: /status
          domainKey: PUBLIC_DOMAIN