AI Actions

Generating Text

What it does: Generates text from a prompt using an AI model.

Inputs

• Prompt (required) — The instruction or question for the model. Can include variables (e.g. #StepName.field).
• Model (required) — Which AI model to use (e.g. GPT-4, Claude). Dropdown or list from your plan.
• Temperature (optional) — Controls randomness (0 = deterministic, higher = more creative). Typical range 0–1.
• Max Tokens (optional) — Maximum length of the generated response in tokens. Prevents runaway output.

Outputs

The model’s generated text and a word count so you can use the result in later steps or for logging.

Output schema

• text — The full generated text string.
• wordCount — Number of words in the generated text (if exposed).

Extracting Information

What it does: Extracts structured information from unstructured text using AI.

Inputs

• Content (required) — The source text (or variable containing text) to extract from (e.g. email body, transcript).
• Schema / Field definitions (required) — The structure you want: list of field names and optionally types or descriptions (e.g. “name: string”, “amount: number”). Tells the AI what to extract.

Outputs

A single object whose keys match your schema and values are the extracted data. Use in conditions or to create/update records.

Output schema

• fieldName — One key per field you defined in the schema. Types and nesting depend on your definition (e.g. name, email, amount, date).

Using a Decoder Assistant

What it does: Uses a Decoder Assistant to answer or perform a task (conversational step).

Inputs

• Assistant (required) — Which Decoder Assistant to use (dropdown of configured Assistants).
• Prompt / Query (required) — The question or instruction for the Assistant. Can include variables.
• Conversation context (optional) — Previous messages or context to maintain a thread (if the step supports it).
• Model overrides (optional) — Override the Assistant’s default model for this step, if allowed.

Outputs

The Assistant’s reply text and, if configured, any structured fields (e.g. extracted JSON). Use for follow-up steps or branching.

Output schema

• text / response — The Assistant’s full reply text.
• structuredField (if configured) — Any additional structured output fields defined for the Assistant step.

Generating Structured Data with an Assistant

What it does: Uses an Assistant to produce formatted or structured output (e.g. JSON, table).

Inputs

• Assistant (required) — Which Assistant to use.
• Prompt (required) — Instruction describing what structure you want (e.g. “Return a JSON object with keys: title, summary, bullets”).
• Format (required) — Schema or format description (JSON Schema, table columns, or natural-language description) so the output matches a known shape.

Outputs

A single object or list in the requested format, suitable for mapping to CRM fields or other steps.

Output schema

• data — Object or array matching the format you specified. Use the variable picker to see the exact shape (e.g. #StepName.data.title).

Summarizing Content

What it does: Summarizes provided content using AI.

Inputs

• Content (required) — The text (or variable) to summarize (e.g. long email, transcript, article).
• Length (optional) — Target length (e.g. short, medium, long, or word count). Affects how much the AI condenses.
• Style (optional) — Tone or style (e.g. bullet points, paragraph, executive summary).

Outputs

A shorter summary string you can use in emails, tickets, or reports.

Output schema

• summary / text — The summary text.

Classifying Content

What it does: Classifies content into predefined categories using AI.

Inputs

• Content (required) — The text (or variable) to classify (e.g. support message, feedback).
• Categories (required) — The list of possible categories (e.g. “Bug”, “Feature request”, “Question”). Can be a simple list or a schema with descriptions.

Outputs

The chosen category and optionally confidence or per-category scores for routing or analytics.

Output schema

• category — The selected category label.
• confidence (optional) — Confidence score (e.g. 0–1).
• scores (optional) — Per-category scores or probabilities, if exposed.

Generate Structured Data

What it does: Generates data (e.g. list of objects) that matches a JSON schema.

Inputs

• Prompt / Instructions (required) — Natural-language description of what to generate (e.g. “Generate 5 product ideas with name and one-line description”).
• Schema (required) — JSON Schema defining the structure (object shape, array of objects, types). The model’s output is validated against this.
• Model (optional) — Override the default model for this step.

Outputs

Valid structured data (object or array) matching the schema. Use for creating records or feeding into iterators.

Output schema

• data — The generated object or array. Fields match your JSON Schema (e.g. data[0].name, data[0].description).

Analyzing Images

Decoder provides image analysis steps that return text or structured data from images.

Simple Image Analysis

What it does: Analyzes an image and returns text output (with optional streaming). Uses an AI image model.

Inputs

• Model (required) — Which vision/model to use (e.g. GPT-4 Vision, Claude with vision).
• Image (required) — The image file or URL (variable from file upload or previous step). Supported: PNG, JPEG, GIF, WebP.
• Prompt (required) — What to do with the image (e.g. “Describe this”, “Extract the text”, “List items in the image”).
• Temperature (optional) — Randomness of the text response.
• Streaming (optional) — When enabled, output may stream token-by-token (behavior depends on product).

Outputs

Plain-text analysis or description of the image for use in summaries, tickets, or next steps.

Output schema

• text — The model’s text response about the image.
• wordCount — Word count of the response (if exposed).

Structured Image Analysis

What it does: Analyzes an image with a user-defined output schema and JSON Schema validation.

Inputs

• Model (required) — Vision-capable model to use.
• Image (required) — Image file or URL (variable).
• Prompt (required) — Instruction for what to extract (e.g. “Extract product name, price, and SKU from this label”).
• Output Schema (required) — JSON Schema for the expected output (object with specific fields). Ensures structured, typed results.
• Temperature (optional) — Randomness; often set low for extraction.

Outputs

A single object (or list) matching your schema—e.g. form fields, product details—for use in databases or other steps.

Output schema

• data — Object whose keys match your Output Schema (e.g. productName, price, sku). Use the variable picker to see exact fields.

OCR Document (Structured Data)

What it does: Extracts structured data from documents (PDF, images, Office files) using OCR and a defined schema.

Inputs

• Document (required) — The file (or variable) to process. Supported: PDF, JPEG, PNG, DOCX, PPTX, XLSX, HTML.
• Output Schema (required) — JSON Schema for the extracted data (field names and types). Defines what to look for and return.
• Schema prompt (optional) — Extra instructions for the extractor (e.g. “Ignore headers and footers”).
• Fail if not found (optional) — When enabled, the step fails if required fields cannot be extracted; when disabled, may return partial or null.

Outputs

Structured data matching your schema, suitable for CRM, DB, or conditional logic.

Output schema

• data — Object matching your Output Schema. Keys are your defined field names; values are extracted text or numbers (e.g. invoiceNumber, total, date).