Action References

AiCallAction

Call an LLM (OpenAI, Anthropic, …) with a structured prompt to get text or JSON.Use it to generate labels, summaries, entity extraction, validation suggestions, or decision scaffolds used by later actions.

Purpose Call an LLM (OpenAI, Anthropic, …) with a structured prompt to get text or JSON. Use it to generate labels, summaries, entity extraction, validation suggestions, or decision scaffolds used by later actions.

When to use it (milestones)

afterBuildDataClause (pre-persist checks, computed fields)
afterMainCreateOperation / afterMainUpdateOperation (post-write summaries, notifications)
afterMainGetOperation / afterMainListOperation (enrichment)
afterBuildOutput (final phrasing/formatting)

Key fields

Field	Type	Notes
`promptTemplate`	`AIPromptTemplate`	Defines system and user parts; user supports MScript.
`model`	`String`	Provider’s model ID (e.g., `"gpt-4o"`, `"claude-3-opus"`).
`responseFormat`	`String`	`"text"` or `"json"` — guide downstream handling.
`aiFetchProperty`	`String`	If model returns an object and you want a single property.
`isArray`	`Boolean`	Set `true` if you expect a list of items.
`contextPropertyName`	`String`	Where result is stored (`this.<name>`).

Behavior

Composes the prompt at runtime using AIPromptTemplate.
Awaits response; if responseFormat:"json", the runtime expects valid JSON (handle errors with onActionError or pre-validate with another step).
Writes result to this.<contextPropertyName>; set writeToResponse:true to expose it.

Example — extract product tags as JSON

{
  "id": "a180-ai-extract-tags",
  "extendClassName": "AiCallAction",
  "name": "extractProductTags",
  "model": "gpt-4o",
  "responseFormat": "json",
  "promptTemplate": {
    "systemPrompt": "You are a tagging engine that outputs strict JSON.",
    "userPrompt": "`{ name: ${this.product.name}, desc: ${this.product.description} }`"
  },
  "aiFetchProperty": null,
  "isArray": false,
  "contextPropertyName": "extractedTags",
  "writeToResponse": false
}

AIPromptTemplate

Purpose Separates system and user instructions cleanly. The user portion can be built with MScript using this values (parameters, session, previous actions).

Fields

Field	Type	Description
`systemPrompt`	`Text`	Stable role/instructions (behavioral guardrails).
`userPrompt`	`MScript`	Dynamic content referencing context (e.g., `${this.description}`).

Example (used above)

"promptTemplate": {
  "systemPrompt": "You are a tagging engine that outputs strict JSON.",
  "userPrompt": "`{ name: ${this.product.name}, desc: ${this.product.description} }`"
}

Was this page helpful?

Built with Documentation.AI

Last updated Jan 3, 2026