Flow step types overview

Runtype provides 20+ step types for building Flows. In the Flow builder, step types are grouped under Starters, AI, Context, Actions, and Replies. This overview covers the main categories and when to use each.

AI and LLM steps

Run Task

Call an AI model with a prompt and get a response. The most commonly used step type. You can configure the prompt step for structured output (e.g. JSON schema) so the model returns data in a specific format.

Use for: Text generation, analysis, classification, summarization, question answering, extracting structured data

Run Agent

Conversational AI with message history and tool-calling capabilities.

Use for: Multi-turn conversations, agentic workflows with tools, complex reasoning tasks

Data transformation

Transform Data

Manipulate data using JavaScript (e.g. format data, parse JSON, merge objects).

Use for: Formatting data, calculations, string manipulation, object merging, parsing JSON

Control flow

Conditional

Branch execution based on conditions.

Use for: If/else logic, routing, error handling

HTTP and APIs

Fetch URL

Make HTTP GET requests to external URLs. Can use Firecrawl for web scraping by setting the fetch method.

Use for: Fetching web pages, reading public data, webhook verification, scraping structured data

Make API Call

Make HTTP requests with full control over method, headers, and body.

Use for: REST API integration, webhooks, third-party services

Paginate API

Automatically iterate through paginated API responses.

Use for: Fetching complete datasets from paginated endpoints

Records and data storage

Retrieve Record

Fetch a Record from Runtype's Record store by type, name, or ID. (Appears in the UI as "Retrieve Another Record.")

Use for: Loading customer data, Product catalogs, knowledge bases

Upsert Record

Create or update a Record.

Use for: Saving conversation history, caching results, updating customer data

Update Record

Modify an existing Record's metadata.

Use for: Updating Record fields, changing Record state

Generate Embedding

Create vector embeddings from text using an AI model.

Use for: Preparing data for semantic search, RAG pipelines

Store Vector

Save vector embeddings to a vector store.

Use for: Building searchable embedding databases, RAG systems

Vector Search

Find similar content using vector embeddings (works with your vector store, e.g. Weaviate).

Use for: Semantic search, finding relevant documents, RAG (Retrieval Augmented Generation)

AI Search

Search the web using AI-powered models (e.g. Exa) or model-based search.

Use for: Finding current information, web research, augmenting AI responses with web data

Utilities and actions

Wait Until

Delay execution for a set time and/or poll an API until a condition is met (e.g. status code, body content, custom expression).

Use for: Rate limiting, waiting for external processes, polling until ready, scheduled Flows

Send Email

Send email messages.

Use for: Notifications, alerts, customer communication

Send Event

Emit custom events during Flow execution.

Use for: Logging, analytics, triggering external workflows

Send Stream of Data

Stream data progressively during Flow execution.

Use for: Real-time updates, streaming responses to clients

Tool Call

Invoke a custom tool or external function.

Use for: Calling agent tools, integrating custom logic

Working with external services

Some steps use external services as providers or backends:

• AI Search can use Exa or model-based search

• Fetch URL can use Firecrawl for scraping by setting the fetch method

• Vector Search and Store Vector work with your vector store (e.g. Weaviate)

For actions like creating GitHub or Linear issues, or sending Slack messages, use Make API Call or agent tools.

Choosing step types

Most Flows follow this pattern:

1. Input validation — Transform or validate incoming data

2. Data retrieval — Fetch Records or call APIs

3. AI processing — Use prompt steps for generation or analysis

4. Output formatting — Transform results to desired format

5. Side effects — Upsert Records, call webhooks, etc.

Next steps

• Using prompt steps for AI model configuration

• Using transform-data steps for data manipulation

• Using conditional steps for branching logic

• Using fetch-url and api-call steps for HTTP requests