bpm-ai-connector-compose.json

{
  "$schema": "https://unpkg.com/@camunda/zeebe-element-templates-json-schema/resources/schema.json",
  "name": "BPM AI Compose Connector",
  "id": "io.holunda.connector.compose.v1",
  "description": "Compose text for email, chat messages or social media posts, in any language",
  "version": 2,
  "documentationRef": "https://github.com/holunda-io/bpm-ai-connectors-camunda-8",
  "icon": {
    "contents": "data:image/svg+xml;base64,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"
  },
  "category": {
    "id": "connectors",
    "name": "Connectors"
  },
  "appliesTo": ["bpmn:Task"],
  "elementType": {
    "value": "bpmn:ServiceTask"
  },
  "groups": [
    {
      "id": "model",
      "label": "Model"
    },
    {
      "id": "input",
      "label": "Input"
    },
    {
      "id": "write",
      "label": "Compose"
    },
    {
      "id": "output",
      "label": "Output Mapping"
    }
  ],
  "properties": [
    {
      "type": "Hidden",
      "value": "io.holunda:connector-compose:2",
      "binding": {
        "type": "zeebe:taskDefinition:type"
      }
    },
    {
      "id": "llm",
      "label": "LLM / Model",
      "group": "model",
      "type": "Dropdown",
      "value": "gpt-4-turbo",
      "choices": [
        { "name": "Anthropic Claude 3.5 Sonnet (best)", "value": "claude-3-5-sonnet-latest" },
        { "name": "Anthropic Claude 3 Opus", "value": "claude-3-opus-latest" },
        { "name": "Anthropic Claude 3 Haiku (fastest)", "value": "claude-3-haiku-latest" },
        { "name": "OpenAI GPT-4o", "value": "gpt-4o" },
        { "name": "OpenAI GPT-4o mini", "value": "gpt-4o-mini" },
        { "name": "OpenAI GPT-4 Turbo", "value": "gpt-4-turbo" },
        { "name": "OpenAI GPT-4", "value": "gpt-4" },
        { "name": "OpenAI GPT-3.5 Turbo", "value": "gpt-3.5-turbo" },
        { "name": "Azure OpenAI", "value": "azure-openai" },
        { "name": "OpenAI Compatible Server", "value": "openai-compatible" },
        { "name": "Groq Llama 3 70B", "value": "groq-llama3-70b-8192" },
        { "name": "Local LLM", "value": "local-llm" }
      ],
      "binding": {
        "type": "zeebe:input",
        "name": "llm"
      }
    },
        {
      "label": "Local LLM",
      "group": "model",
      "type": "Dropdown",
      "value": "QuantFactory/Phi-3-mini-4k-instruct-GGUF",
      "choices": [
        { "name": "Meta Llama 3.1 8B (local)", "value": "bartowski/Hermes-3-Llama-3.1-8B-GGUF" },
        { "name": "Meta Llama 3.2 3B (local)", "value": "unsloth/Llama-3.2-3B-Instruct-GGUF" },
        { "name": "Meta Llama 3.2 1B (local)", "value": "unsloth/Llama-3.2-1B-Instruct-GGUF" },
        { "name": "Microsoft Phi-3 Mini (local)", "value": "QuantFactory/Phi-3-mini-4k-instruct-GGUF" },
        { "name": "Mistral 7B (local)", "value": "NousResearch/Hermes-2-Pro-Mistral-7B-GGUF" }
      ],
      "binding": {
        "type": "zeebe:input",
        "name": "local_llm"
      },
      "condition": {
        "property": "llm",
        "equals": "local-llm"
      }
    },
    {
      "label": "LLM Precision",
      "group": "model",
      "type": "Dropdown",
      "value": "*Q5_K_M.gguf",
      "choices": [
        { "name": "Low (smallest & fastest)", "value": "*Q2_K.gguf" },
        { "name": "Balanced", "value": "*Q5_K_M.gguf" },
        { "name": "High (largest & slowest)", "value": "*Q8_0.gguf" }
      ],
      "binding": {
        "type": "zeebe:input",
        "name": "model_filename"
      },
      "condition": {
        "property": "llm",
        "equals": "local-llm"
      }
    },
    {
      "label": "Endpoint",
      "group": "model",
      "type": "String",
      "value": "",
      "feel": "optional",
      "binding": {
        "type": "zeebe:input",
        "name": "model_endpoint"
      },
      "condition": {
        "property": "llm",
        "oneOf": ["azure-openai", "openai-compatible"]
      }
    },
    {
      "label": "OCR Model",
      "group": "model",
      "type": "Dropdown",
      "value": "none",
      "choices": [
        { "name": "None", "value": "none" },
        { "name": "Amazon Textract", "value": "amazon-textract" },
        { "name": "Azure Document Intelligence", "value": "azure-document-intelligence" },
        { "name": "Tesseract (local)", "value": "tesseract" }
      ],
      "binding": {
        "type": "zeebe:input",
        "name": "ocr"
      }
    },
    {
      "label": "Speech Recognition Model",
      "group": "model",
      "type": "Dropdown",
      "value": "openai-whisper-1",
      "choices": [
        { "name": "None", "value": "none" },
        { "name": "OpenAI Whisper API", "value": "openai-whisper-1" },
        { "name": "OpenAI Whisper Small (local)", "value": "faster_whisper-small" },
        { "name": "OpenAI Whisper Medium (local)", "value": "faster_whisper-medium" },
        { "name": "OpenAI Whisper Large (local)", "value": "faster_whisper-large" }
      ],
      "binding": {
        "type": "zeebe:input",
        "name": "asr"
      }
    },
    {
      "label": "Input Variables",
      "group": "input",
      "description": "Map of input variables to consider for the text. Key should be the variable name. Value is the variable itself as FEEL expression.",
      "type": "Text",
      "feel": "optional",
      "binding": {
        "type": "zeebe:input",
        "name": "input_json"
      },
      "value": "={ \"myVariable\": myVariable }",
      "constraints": {
        "notEmpty": true
      }
    },
    {
      "id": "type",
      "label": "Type",
      "group": "write",
      "type": "Dropdown",
      "value": "letter",
      "choices": [
        { "name": "E-Mail/Letter", "value": "letter" },
        { "name": "Chat Message", "value": "chat" },
        { "name": "Social Media Post", "value": "social" },
        { "name": "General Text", "value": "text" }
      ],
      "binding": {
        "type": "zeebe:input",
        "name": "properties.type"
      }
    },
    {
      "label": "Style",
      "group": "write",
      "type": "Dropdown",
      "value": "formal",
      "choices": [
        { "name": "Formal", "value": "formal" },
        { "name": "Informal", "value": "informal" }
      ],
      "binding": {
        "type": "zeebe:input",
        "name": "properties.style"
      }
    },
    {
      "label": "Tone",
      "group": "write",
      "type": "Dropdown",
      "value": "friendly",
      "choices": [
        { "name": "Friendly", "value": "friendly" },
        { "name": "Overly Friendly", "value": "overly friendly" },
        { "name": "Neutral", "value": "neutral" },
        { "name": "Assertive", "value": "assertive" }
      ],
      "binding": {
        "type": "zeebe:input",
        "name": "properties.tone"
      }
    },
    {
      "label": "Language",
      "group": "write",
      "type": "String",
      "feel": "optional",
      "value": "English",
      "binding": {
        "type": "zeebe:input",
        "name": "properties.language"
      },
      "constraints": {
        "notEmpty": true
      }
    },
    {
      "label": "Length",
      "group": "write",
      "type": "Dropdown",
      "value": "adequate",
      "choices": [
        { "name": "Very Brief", "value": "very brief" },
        { "name": "Brief", "value": "brief" },
        { "name": "Adequate", "value": "adequate" },
        { "name": "Lengthy", "value": "lengthy" }
      ],
      "binding": {
        "type": "zeebe:input",
        "name": "properties.length"
      }
    },
    {
      "label": "Variance",
      "group": "write",
      "description": "<code>High</code> results in more diverse and unpredictable text, <code>None</code> makes the output focused and deterministic.",
      "type": "Dropdown",
      "value": "0.0",
      "choices": [
        { "name": "None", "value": "0.0" },
        { "name": "Low", "value": "0.2" },
        { "name": "Medium", "value": "0.5" },
        { "name": "High", "value": "0.7" }
      ],
      "binding": {
        "type": "zeebe:input",
        "name": "properties.temperature"
      }
    },
    {
      "label": "Template",
      "description": "Template variables found in input variables are replaced directly (but not sent to the model), text for other variables will be generated.",
      "group": "write",
      "type": "Text",
      "feel": "optional",
      "value": "Hello {customerName},\n\n{ thank the customer for his purchase }\n\nYours,\n{companyName}",
      "binding": {
        "type": "zeebe:input",
        "name": "template"
      },
      "constraints": {
        "notEmpty": true
      }
    },
    {
      "label": "Result Expression",
      "description": "Expression to map the result into process variables. Result text is in a temporary variable <code>result.text</code>.",
      "group": "output",
      "type": "Text",
      "feel": "required",
      "value": "={\"text\": result.text}",
      "binding": {
        "type": "zeebe:taskHeader",
        "key": "result_expression"
      }
    }
  ]
}