Runtype
Tools

Built-in tools

Runtype provides seven built-in tools that Agents and Flows can use: image generation (DALL-E), web search (OpenAI Web Search, Anthropic Web Search, Exa), web scraping (Firecrawl), web content fetch (Anthropic Web Fetch), and semantic search over knowledge bases (Semantic Search). You add them when configuring a prompt step or an Agent by choosing a compatible model; configure the required API keys in Settings as needed.

How built-in tools work

Built-in tools appear in the tool picker when you add tools to a Flow prompt step or an Agent and select a model that supports that tool. Only tools compatible with the selected model are shown.

To use them:

  1. Go to Settings → Integrations to add API keys for Exa and Firecrawl (or use Runtype platform keys where offered).

  2. For DALL-E and provider web search tools, ensure the right provider is configured (e.g. Settings → Models for OpenAI or Anthropic).

  3. When editing a prompt step or an Agent, open the tools configuration, choose Built-in Tools, and add the tools you want.

The tool is then available for that prompt or Agent.

DALL-E (Image generation)

Generate images from text descriptions using OpenAI's DALL-E model. Available when you select an OpenAI tool-calling model (e.g., gpt-4o, gpt-4o-mini, gpt-4-turbo).

Configuration

  • Requires OpenAI API key (configured in Settings → Models)

  • Or use Runtype platform key (rate-limited)

Parameters

  • prompt: Description of image to generate (required, 1–4000 characters)

  • size: Image dimensions (1024x1024, 1792x1024, 1024x1792; default: 1024x1024)

  • quality: Standard or hd (default: standard)

  • style: Vivid or natural (default: vivid)

Example use

Agent prompt: "Create a logo for a coffee shop"

Agent calls DALL-E with prompt, size, quality, and style parameters.

Returns: URL to generated image

Native web search for compatible OpenAI models. Searches the web and returns current information.

Configuration

  • Requires OpenAI API key (configured in Settings → Models)

  • Available for OpenAI tool-calling models

Parameters

  • searchContextSize: Low, medium, or high (default: medium)

  • userLocationType: None, approx, or exact (default: none)

  • userLocationCity, userLocationRegion: Optional location strings

Native web search for compatible Anthropic Claude models. Searches the web and returns current information.

Configuration

  • Requires Anthropic API key (configured in Settings → Models)

  • Available for Claude models (e.g., claude-3-5-sonnet)

Parameters

  • maxUses: Maximum number of search requests (1–10; default: 5)

Anthropic Web Fetch

Fetch and extract content from URLs for compatible Anthropic Claude models.

Configuration

  • Requires Anthropic API key (configured in Settings → Models)

  • Available for Claude models

Parameters

  • maxUses: Maximum number of fetch requests (default: 1)

Search the web for current information using Exa's AI-powered search. Available across model providers.

Configuration

  1. Sign up for Exa at exa.ai

  2. Get your API key

  3. Add to Settings → Integrations (or use Runtype platform key)

Parameters

  • query: Search query (required)

  • searchType: auto, neural, or keyword (default: auto)

  • numResults: How many results to return (1–10; default: 5)

  • text, summary, highlights: Optional booleans to include full text, summary, or highlights

  • includeDomains, excludeDomains: Optional arrays to filter results

Example use

Agent prompt: "Find recent news about quantum computing breakthroughs"

Agent calls Exa with query, searchType, and numResults parameters.

Returns: Array of search results with titles, URLs, and snippets

Firecrawl (Web scraping)

Extract content from web pages with Firecrawl. Available across model providers.

Configuration

  1. Sign up for Firecrawl at firecrawl.dev

  2. Get your API key

  3. Add to Settings → Integrations (or use Runtype platform key)

Parameters

  • url: URL to scrape (required)

  • formats: Content formats to extract—markdown, html, rawHtml, screenshot, links (default: [markdown])

  • onlyMainContent: Whether to extract only main content (default: true)

  • waitFor: Optional delay in milliseconds before extracting (0–30000; for dynamic pages)

Example use

Agent prompt: "Get the main content from example.com/article"

Agent calls Firecrawl with url, formats, and onlyMainContent parameters.

Returns: Extracted content in the specified formats

Search over knowledge bases using vector search. You can configure multiple instances with different knowledge bases and vector providers.

Configuration

In the Agent or Flow tools configuration, add a knowledge base instance:

  1. Click Add Knowledge Base

  2. Set a display name

  3. Choose a Vector Provider: Platform pgvector, Weaviate, or Cloudflare Vectorize

  4. Configure provider-specific settings (e.g., record type for pgvector, collection name for Weaviate, index name for Vectorize)

  5. Set similarity threshold (0–1 slider)

Semantic Search uses OpenAI embeddings, so you'll need an OpenAI API key configured in Settings → Models.

Parameters

  • query: Search query (required)

  • limit: Maximum number of results (1–20; default: 5)

  • threshold: Similarity threshold (0–1; default: 0.3)

Example use

Agent searches a configured knowledge base for relevant documents matching a query.

Returns: Array of matching records with metadata and similarity scores

Built-in tools use your API keys when configured. If no key is set, some tools fall back to Runtype's platform keys with rate limits.

Adding built-in tools to Agents and Flows

For Agents:

  1. Open your Agent

  2. In the Capabilities tab, go to the Tools section and click Configure Tools

  3. In the Tool Selection sheet, select from Built-in Tools

  4. Choose the tools you want (only compatible tools for the selected model appear)

  5. Click Apply Changes

For Flows:

  1. Open your Flow

  2. Add a prompt step or open an existing one

  3. In the tools configuration, choose Built-in Tools

  4. Select the tools you want and configure their parameters

Which built-in tools appear depends on the model selected. For example, DALL-E and OpenAI Web Search are available for OpenAI models; Anthropic Web Search and Web Fetch are available for Claude models; Exa, Firecrawl, and Semantic Search are available across model providers.

Cost considerations

Built-in tools consume API credits from their respective providers:

  • DALL-E: ~$0.04 per image (standard quality)

  • Exa: Varies by plan

  • Firecrawl: Varies by plan

  • Semantic Search: OpenAI embedding costs per query

Monitor usage in each provider's dashboard.

Best practices

  • Connect your own API keys for high volume: BYOK provides dedicated rate limits and separate cost tracking

  • Descriptive tool use: Give Agents clear instructions about when to use each tool

  • Limit results: Don't fetch more data than needed (e.g., numResults: 3 vs 10)

  • Monitor costs: Track API usage to avoid surprise bills

Was this helpful?