Built-in Tools

Runtype includes built-in tools that you can add to any Agent or Flow. These tools let your AI generate images, search the web, fetch page content, scrape sites, browse pages, manage records, run sandboxed code, search your knowledge bases, remember information across sessions, and work with dates and times without writing custom code.

The core tools are:

Additional built-in tool categories include Browser tools (look at and interact with live web pages), File Operations (store assets, publish pages, generate PDFs), Record Management (CRUD on Runtype records), Agent Memory (persist facts and recall them across sessions), Sandbox (run code in persistent Linux containers), Send Email (transactional email via Resend), and Temporal tools (get the current time, convert between timezones, remember a user’s timezone, and schedule agent reminders).

Tools labeled “Any model” work regardless of which AI provider you choose, which makes them a strong default for multi-model workflows.

Temporal tools

AI models have no internal clock and no sense of “now” beyond their training cutoff. Temporal tools ground your Agent in real-world time. They work with any model and need no API key.

Timezone resolves with a layered precedence: an explicit timezone passed to a tool wins, then the conversation’s stored timezone (set by Set Timezone), then the Agent’s default, then UTC. For example, when a user mentions where they are, the Agent can call Set Timezone with the matching IANA name (such as “Asia/Tokyo”), and every later Get Current Time call renders in their local zone. The conversation timezone can also be set programmatically through the messaging conversations API. Set the Agent’s default timezone through its temporal configuration.

Set Reminder

Set Reminder lets an Agent schedule itself to run again at a future time — a durable “wake me later”. It creates a one-time Schedule targeting the running Agent, so when it fires the Agent receives the reminder as its kickoff message and can act on it (including reaching out to the user through its messaging or email tools).

Set Reminder is available only to Agents. Reminders count toward your plan’s active schedule limit, the same as schedules created from the dashboard or API.

Getting started

Most built-in tools work out of the box with Runtype platform keys, so you can try them right away. For higher-volume usage, connect your own API keys.

• OpenAI tools (GPT Image 2, Web Search): Go to Settings → Models and add your OpenAI API key, or use the included platform key. See Connecting AI model providers.
• Anthropic tools (Web Search, Web Fetch): Go to Settings → Models and add your Anthropic API key.
• Exa: Go to Settings → Integrations and add your Exa API key, or use the included platform key.
• Firecrawl: Go to Settings → Integrations and add your Firecrawl API key, or use the included platform key.
• Massive: Go to Settings → Integrations and add your Massive API token, or use the included platform key.
• Semantic Search: Requires an OpenAI API key for embeddings in Settings → Models. If you need background on vector data, see What are Records?.

Adding tools to an Agent

1. Open your Agent and go to the Capabilities tab.
2. In the Tools section, click Configure Tools.
3. Select tools from the Built-in Tools category. Only tools that match your Agent’s model will appear.
4. Click Apply Changes.

For the full Agent setup flow, see Creating and configuring Agents.

Adding tools to a Flow

1. Open your Flow and add or edit a prompt step.
2. Open the tools configuration and choose Built-in Tools.
3. Select the tools you need and configure any optional settings.

For a step-by-step Flow walkthrough, see Creating and Editing Flows.

If a tool does not appear, check the model on the prompt step or Agent. Some tools are provider-specific. You can also review Available AI models.

GPT Image 2

GPT Image 2 generates images from text descriptions using OpenAI’s image model. Use it to create visuals, mockups, logos, and illustrations inside a Flow or Agent conversation.

Settings

• Size: Default is 1024x1024. Supports auto and custom WxH resolutions where both edges are multiples of 16 (max 3840px per edge).
• Quality: low (fast, ~10s), medium, or high (detailed, may take 1-3+ minutes). Default is low.
• Output format: png, jpeg, or webp
• Background: auto or opaque

Example: An Agent can answer “Create a logo for a coffee shop” by generating and returning an image URL.

OpenAI Web Search

OpenAI Web Search lets OpenAI models search the web and return current information during a conversation.

Settings

• Search context size: Low, medium, or high
• Location context: Optionally provide a city and region for location-specific results

Pricing: When you use a Runtype platform key, each web search is billed a small per-search fee on top of the model’s token costs. Connect your own OpenAI API key to pay OpenAI directly and avoid the platform fee. Search fees appear under Spend by tool in your usage data.

Anthropic Web Search

Anthropic Web Search lets Claude models search the web for current information.

Settings

• Max uses: Limit how many searches the model can make per response from 1 to 10. The default is 5.

Pricing: When you use a Runtype platform key, each web search is billed a small per-search fee on top of the model’s token costs. Connect your own Anthropic API key to pay Anthropic directly and avoid the platform fee. Search fees appear under Spend by tool in your usage data.

Anthropic Web Fetch

Anthropic Web Fetch lets Claude Sonnet models fetch and read content from specific URLs. Use it when your prompt references a known page and you want the model to pull that content directly.

Settings

• Max uses: Limit how many pages the model can fetch per response from 1 to 10. The default is 1.

Exa

Exa provides AI-powered web search with semantic and keyword options. It works with any model and supports filtering for research-heavy workflows.

Settings

• Search type: Auto, neural, or keyword
• Number of results: 1 to 10. The default is 5.
• Content options: Include full text, AI-generated summaries, or highlights
• Domain filtering: Restrict or exclude specific domains
• Date filtering: Limit results to a date range

Example: An Agent can search for “recent breakthroughs in quantum computing” and return titles, URLs, and summaries.

Use domain filtering to keep results focused. For example, restrict results to documentation sites when you are building a support Agent.

Firecrawl

Firecrawl extracts clean, structured content from web pages. It handles JavaScript-rendered pages and removes navigation, ads, and footers so your AI gets the main content.

Settings

• Output formats: Markdown, HTML, raw HTML, screenshot, or links
• Main content only: Enabled by default to remove sidebars, navigation, and footers
• Wait time: Add up to 30 seconds for pages with dynamic content

Example: An Agent can scrape a product page and extract the main content as Markdown for summarization or analysis.

Massive Web Render

Massive fetches hard-to-retrieve pages through its Web Render API. It renders JavaScript, solves captchas, and retries through a residential network, which makes it a good fit for sites that block or challenge datacenter traffic. See Massive (web rendering) for setup details.

Settings

• Output format: Markdown (default), rendered HTML, or raw HTML
• Country: Two-letter ISO code to fetch from a specific country (for example us)
• Delay: 0.1 to 10 supplemental seconds before capturing content on slow-loading pages

Example: An Agent can fetch a page that returns a 403 to ordinary requests and read it as Markdown, optionally as seen from a specific country.

Semantic Search

Semantic Search lets you search knowledge bases using vector similarity. Instead of matching exact keywords, it finds content by meaning.

You can configure multiple Semantic Search instances on a single Agent or Flow step, with each instance pointing to a different knowledge base.

Set up a knowledge base instance

1. In the tool configuration, click Add Knowledge Base.
2. Enter a display name. This is what the AI model sees, such as “Search Documentation.”
3. Choose a vector provider.
4. Set the similarity threshold from 0 to 1. Lower values return more results. Higher values return only closer matches.

Settings

• Limit: Maximum number of results per query from 1 to 20. The default is 5.

Example: A support Agent can search a product documentation knowledge base before responding to a customer question.

Start with a similarity threshold around 0.3 and adjust based on result quality. If the threshold is too high, you may miss relevant content. If it is too low, you may get noisy results.

Agent Memory tools

Agent Memory tools let an Agent remember what it learns and recall it in later conversations. Unlike Semantic Search, which queries knowledge bases you load ahead of time, Agent Memory is the Agent’s own long-term store: facts, preferences, decisions, and events it picks up while talking to a user, kept across sessions. They work with any model and need no API key.

The tools come as a bundle and become available when memory is enabled on the Agent. To turn them on, set memory: { enabled: true } in the Agent configuration — see Memory for the full reference, including profileTemplate for per-user sharding.

Recall from Memory accepts an optional thinkingLevel (low, medium, high) to control search breadth and an optional responseLength (short, medium, long) to control how detailed the synthesized answer is.

Example: A support Agent saves a customer’s plan tier and timezone the first time they come up, then recalls them in a later conversation instead of asking again.

When memory is not enabled for an Agent, the memory tools return a clear memory_not_enabled result rather than failing silently.

Browser tools

Browser tools let an Agent or Flow look at and interact with live web pages. They appear under a dedicated Browser category in the tool picker and cover three styles of use: quick one-shot actions, persistent browsing sessions, and an autonomous browsing agent.

Quick-action tools

Quick-action tools are stateless and run against a single URL without keeping a session open. Use them when you need to grab content from one page and move on.

• browser:screenshot: Capture a screenshot of a rendered page.
• browser:content: Get the fully rendered HTML after JavaScript runs.
• browser:markdown: Convert the page into clean Markdown.
• browser:scrape: Extract data from specific CSS selectors.
• browser:json: Extract structured JSON from the page.
• browser:links: Retrieve the links on a page.
• browser:snapshot: Capture rendered HTML and a screenshot together.

Session tools

Session tools share a persistent browser session so the AI can take several steps on the same page. A session keeps its state across multiple turns in the same conversation, so an Agent can open a page on one turn and continue interacting with it on the next.

• browser:open: Start a session and load a page.
• browser:navigate: Go to a new URL in the session.
• browser:click: Click an element.
• browser:type: Type into an input.
• browser:scroll: Scroll the page.
• browser:session-screenshot: Capture the current session view.
• browser:evaluate: Run a script against the page.
• browser:session-info: Read the current session state.
• browser:close: End the session.

Sessions close automatically after 2 minutes of inactivity and have a hard cap of 10 minutes, so finish multi-step work within that window.

Autonomous browsing with browser:agent

browser:agent accepts a natural-language goal and drives the page on its own to accomplish it, rather than requiring you to script each click and type. Give it a goal such as “find the pricing page and report the starting price.”

• Start fresh: Provide a startUrl to begin a new session at a given page.
• Attach to a session: Provide a sessionId to continue inside an existing session.
• Choose the output: It returns a text summary by default, or structured JSON when you provide an outputSchema.

Billing: Session tools are billed at a flat per-action cost. browser:agent is billed dynamically based on usage, with your account margin applied (default 20%).

File operations

File operations tools let an Agent or Flow save files to permanent, shareable URLs. They appear under the File Operations category in the tool picker.

store_asset

store_asset downloads a file from a URL and re-hosts it on Runtype storage, returning a permanent link. This solves the expiring-URL problem: many generated files — AI images especially — come with temporary links that stop working within an hour. Store the file once and share the lasting URL in emails, Records, Slack messages, or product listings.

It returns the permanent (or signed) url, an assetId, the contentType, and sizeBytes. Files are limited to 25 MB. Source URLs are checked before download, and file types that could be unsafe to render in a browser (such as HTML or SVG) are served as downloads rather than displayed inline.

Automatic image hosting

When the image generation tool creates an image, Runtype automatically stores a permanent copy so the returned URL does not expire. The response includes both the permanent hosted URL, which is recommended, and the original temporary URL as a fallback. If hosting fails, the original URL is returned and execution continues.

Controlling who sees media

When a tool produces media — a generated image, for example — you can control who that media reaches. Set the tool’s media visibility to one of three values:

Use model when an image is an intermediate result the AI should reason about but the user doesn’t need to see. Use user when the media is for display only and would just consume the model’s context. The default is both, so tools behave the same way unless you change this setting.

Using platform keys and your own API keys

Runtype provides platform keys for several tools so you can start building without signing up for separate services. Platform keys are shared and rate-limited, which works well for development and lower-volume use.

For production or heavier usage, connect your own API keys in Settings to get dedicated rate limits and full control over usage. To understand the tradeoffs, see Platform Keys vs. Bring Your Own Key (BYOK).

Best practices

• Be specific in your prompts: Tell your Agent when and how to use each tool.
• Limit result counts: Fewer results often keep responses faster and more focused.
• Combine tools strategically: Pair Exa with Firecrawl for end-to-end research, or pair Semantic Search with a prompt step for retrieval workflows.
• Use multiple knowledge bases: Configure separate Semantic Search instances for different data sources, such as documentation and customer Records.

Next steps

• What are Tools? for a broader overview of how built-in, custom, and external tools fit together.
• Creating and configuring Agents if you want to add built-in tools to an Agent.
• Creating and Editing Flows if you want to use built-in tools inside a Flow.
• Connecting AI model providers if you need to add provider keys before tools appear.