AI & Machine Learning Modules

All AI modules use the component system — configure which AI provider to use at the installation level or override per flowlet via componentConfigIds.

Chat — LLM Completion

Sends a prompt to a language model and returns the response.

appId: chat

{
  "appId": "chat",
  "ref": "summarize",
  "args": {
    "systemPrompt": "You are a document summarizer. Be concise and accurate.",
    "userMessage": "Summarize the following invoice:\n\n{{ $prev.extractText.text }}",
    "model": "gpt-4o",
    "temperature": 0.3,
    "maxTokens": 500,
    "history": []
  }
}

Arguments:

Arg	Type	Description
`systemPrompt`	string	System-level instructions for the model
`userMessage`	string	The user’s input message
`model`	string	Model name (provider-specific, e.g., `gpt-4o`, `claude-3-5-sonnet-20241022`)
`temperature`	float	Creativity: 0.0 (deterministic) to 2.0 (random). Default: 0.7
`maxTokens`	int	Maximum tokens in the response
`history`	array	Prior conversation messages `[{"role":"user","content":"..."}]`

Output: { "response": "This invoice from ACME Corp totals $1,250...", "tokens": 142 }

Embeddings — Vector Generation

Converts text into a numerical vector for semantic search and RAG pipelines.

appId: embeddings

{
  "appId": "embeddings",
  "ref": "vectorize",
  "args": {
    "input": "{{ $prev.extractText.text }}",
    "model": "text-embedding-3-small"
  }
}

Output: { "vector": [0.012, -0.034, 0.091, ...], "dimensions": 1536 }

Structured Extract — Schema-Validated Output

Prompts the AI to return data in a specific JSON Schema structure. Useful for extracting structured fields from unstructured text.

appId: structured-extract

{
  "appId": "structured-extract",
  "ref": "extractInvoiceData",
  "args": {
    "input": "{{ $prev.ocrText.text }}",
    "prompt": "Extract invoice data from the following text:",
    "schema": {
      "type": "object",
      "properties": {
        "invoiceNumber": { "type": "string" },
        "totalAmount": { "type": "number" },
        "dueDate": { "type": "string", "format": "date" },
        "vendorName": { "type": "string" },
        "lineItems": {
          "type": "array",
          "items": {
            "type": "object",
            "properties": {
              "description": { "type": "string" },
              "amount": { "type": "number" }
            }
          }
        }
      },
      "required": ["invoiceNumber", "totalAmount", "vendorName"]
    }
  }
}

Output: The data field contains the JSON-validated extracted object. The valid field indicates schema compliance.

AI Prompt — Template Builder

Builds a prompt string from a template with injected data — useful as a precursor step to the Chat module.

appId: ai-prompt

{
  "appId": "ai-prompt",
  "ref": "buildPrompt",
  "args": {
    "template": "Analyze the following {{ .documentType }} and identify any issues:\n\n{{ .content }}\n\nFocus on: {{ .focusArea }}",
    "data": {
      "documentType": "contract",
      "content": "{{ $prev.extractText.text }}",
      "focusArea": "payment terms and liability clauses"
    }
  }
}

Output: { "prompt": "Analyze the following contract and identify..." }

MCP Agent Loop — Tool-Using AI Agent

Runs an AI agent that can use flow8’s MCP tools in a loop until it completes the task.

appId: mcp-agent-loop

{
  "appId": "mcp-agent-loop",
  "ref": "runAgent",
  "args": {
    "task": "Find all overdue invoices from Clio and create a summary report.",
    "tools": ["clio-list-matters", "clio-get-contact", "kv-set"],
    "maxIterations": 10,
    "model": "claude-3-5-sonnet-20241022"
  }
}

Output: { "result": "Found 7 overdue invoices totaling...", "iterations": 4, "toolCallCount": 12 }