AI Actions

AI actions hand work to a model: chat, image generation, image recognition, moderation, and speech conversion. Each one runs against a model you choose and writes its result back into the context.

Model and provider selection

Every AI action carries a model selection that names the provider and model to use for that call. The available providers and models, and how selection is resolved, are covered on AI Models & Providers. Inputs to these actions are usually value inputs (a constant, a context key, or a variable); outputs are written to a named context key.

AI Chat

Sends a prompt to a chat/completion model and stores the text reply. Use it for classification, extraction, summarization, drafting, and any natural-language reasoning step.

Field	Role
System Message	Input — optional instructions that set the model's behavior.
User Message	Input — the main prompt sent to the model.
Output Text	Output — context key that receives the model's reply.

AI Image Generation

Generates an image from a text description and stores the result for later use.

Field	Role
Prompt	Input — a description of the image to generate.
Output	Output — context key that receives the generated image.

AI Image Recognition

Asks a vision-capable model a question about one or more images and returns a text answer.

Field	Role
Images	Input — one or more images to analyze.
Prompt	Input — what to ask the model about the image(s).
Output Text	Output — context key that receives the answer.

An image given as a URL must be on the public internet. With Google Gemini the engine downloads it itself, so it must also answer within 30 seconds, be no larger than 20 MB, and be served directly rather than through a redirect.

AI Moderation

Evaluates text against content policy and reports whether it should be flagged. Use it to gate user-supplied content before acting on it. Moderation runs on OpenAI; see supported providers.

Field	Role
Text	Input — the text to moderate.
Flagged Output	Output — context key that receives the flagged result.

Speech-to-Text

Transcribes audio into text.

Field	Role
Audio	Input — context key holding the audio to transcribe.
Prompt	Input — optional hint to guide transcription.
Output	Output — context key that receives the transcript.

Text-to-Speech

Synthesizes spoken audio from text. Text-to-speech runs on OpenAI; see supported providers.

Field	Role
Text	Input — the text to speak.
Speech Output	Output — context key that receives the generated audio.

Chaining AI steps Because outputs land in context keys, you can pipe one AI action into another — for example, transcribe audio with Speech-to-Text, then feed the transcript into AI Chat for summarization.