Workflow Step Types

Hubrix workflows support 10 step types. Each step does one thing well and passes its output to the next step as {{steps.step_name.output}}.

Prompt

Run an AI prompt with a chosen model.

Required config:

• Model — which AI model to use (GPT-4o, Claude, Gemini, etc.)
• System message — the AI's instructions for this step
• User prompt — the message, optionally using {{variable}} references to earlier steps

Example use case: Summarise a news feed fetched by an HTTP Request step.

Agent Call

Invoke one of your Custom Agents with an input message.

Required config:

• Agent — select from your existing agents
• Input — the message to send to the agent, can reference previous steps

Example use case: Route each row of a dataset through your Data Analyst agent for interpretation.

HTTP Request

Make a GET, POST, or PUT request to any external URL.

Required config:

• Method — GET, POST, or PUT
• URL — the endpoint, with optional {{variable}} interpolation
• Headers — key/value pairs (for example, Authorization: Bearer {{inputs.api_key}})
• Body — for POST/PUT, the request body (JSON template with variable references)

Example use case: Fetch a CRM contact's details before enriching with AI.

Extract

Pull structured fields from unstructured text using AI.

Required config:

• Input text — the text to extract from (usually a reference to a previous step)
• Fields to extract — define each field name and a description of what to extract

Output: A JSON object with each extracted field as a key.

Example use case: Extract invoice number, vendor name, and total amount from a scanned invoice text.

Branch

Route the workflow to different paths based on a condition.

Required config:

• Condition — a variable or expression to evaluate
• True path — steps to run if the condition is true
• False path — steps to run if the condition is false

Example use case: If a lead's score is above 80, send to the sales team via email; otherwise, add to a nurture sequence.

Loop

Iterate over an array, running a set of sub-steps for each item.

Required config:

• Array — the input array to loop over (reference a previous step's array output)
• Sub-steps — the steps to run for each item; use {{item}} to reference the current element

Example use case: For each article in an RSS feed, generate a short summary using a Prompt step.

RAG Query

Search your document library with a query and return the most relevant chunks.

Required config:

• Query — the search question, can reference previous step outputs
• Top K — how many chunks to return (default: 5)
• Document filter — optionally restrict the search to specific documents or folders

Output: An array of text chunks with their source document names.

Example use case: Before drafting an email response, retrieve relevant policy sections from the company knowledge base.

Transform

Reshape or manipulate JSON data using a JavaScript expression.

Required config:

• Input — the data to transform (reference a previous step)
• Expression — a JavaScript expression that returns the transformed output (for example, input.items.map(i => i.name))

Example use case: Extract just the email field from an array of contact objects before passing them to an Email Send step.

Human Input

Pause the workflow and wait for a human to provide input or approval before continuing.

Required config:

• Prompt — the question or instruction shown to the assigned reviewer
• Assignee — which team member receives the approval request (optional; defaults to workflow owner)
• Timeout — how long to wait before marking the run as timed out (max 72 hours)

Example use case: After an AI drafts a client proposal, pause and ask the account manager to approve it before sending.

Email Send

Send a transactional email via the Hubrix email service.

Required config:

• To — recipient email address (can reference a previous step variable)
• Subject — email subject line
• Body — email body, supports HTML and Markdown, can reference previous step outputs

Example use case: Send a finished weekly report to a distribution list every Monday morning.