AI: orchestration

AI orchestration ties the three together: the engine detects the moment in user state, picks the cohort, and chooses the action, so the loop runs without you wiring each step by hand.

• Detects the meaningful moment in live state
• Selects the right cohort and the right action
• Closes the loop: the action becomes the next event the engine learns from

Built-in AI and the AI add-on

Prodantix has two kinds of AI. Built-in AI is part of every plan and always on. It reads each message as it arrives and picks between fixed answers: whether it is spam, whether your knowledge can answer it, and whether the visitor wants a person. It never writes, and it is never billed by the token. The AI add-on writes: Ask, reply drafts and translations, and the messenger assistant's answers. It is off until you turn it on, and it is billed by the token at the model's price plus 20%.

How it works

Every answer is grounded in rows the caller could already read. Citations are resolved server-side from a built candidate list rather than taken from the model’s reply, so an answer can only point at something that exists and the caller is entitled to see.

Where to reach it

Example

query Explain($projectId: String!) {
  ai {
    explainMetric(
      projectId: $projectId
      metric: "order.completed"
      window: "P7D"
    ) {
      summary
      citations { kind id }
    }
  }
}

Pointing an agent at prodantix

The MCP surface exposes the same capabilities to an autonomous agent. The agent signs in with Bomdisoft and you choose what it may do; see MCP for connecting one.

{
  "mcpServers": {
    "prodantix": {
      "url": "https://eu.mcp.prodantix.com/mcp"
    }
  }
}