> ## Documentation Index
> Fetch the complete documentation index at: https://actionbook.dev/docs/llms.txt
> Use this file to discover all available pages before exploring further.

# JavaScript SDK

> Integrate Actionbook into your custom AI agents

<Note>
  The recommended way to get started is the [CLI](/guides/installation). The
  JavaScript SDK is an **optional, advanced** path for developers building
  custom AI agents programmatically.
</Note>

Use this option to integrate Actionbook directly into your custom AI agents built with any LLM framework.

## Integration Examples

<AccordionGroup>
  <Accordion title="With Vercel AI SDK">
    ```typescript theme={null}
    import { Actionbook } from '@actionbookdev/sdk'
    import { generateText, tool } from 'ai'
    import { openai } from '@ai-sdk/openai'

    const actionbook = new Actionbook()

    const { text } = await generateText({
      model: openai('gpt-4o'),
      tools: {
        searchActions: tool({
          description: actionbook.searchActions.description,
          parameters: actionbook.searchActions.params.zod,
          execute: async ({ query }) => actionbook.searchActions(query),
        }),
        getActionById: tool({
          description: actionbook.getActionById.description,
          parameters: actionbook.getActionById.params.zod,
          execute: async ({ id }) => actionbook.getActionById(id),
        }),
      },
      maxSteps: 5,
      prompt: 'Search for LinkedIn message actions and get the action manual',
    })
    ```
  </Accordion>

  <Accordion title="With OpenAI SDK">
    ```typescript theme={null}
    import { Actionbook } from '@actionbookdev/sdk'
    import OpenAI from 'openai'

    const actionbook = new Actionbook()
    const openai = new OpenAI()

    const tools: OpenAI.ChatCompletionTool[] = [
      {
        type: 'function',
        function: {
          name: 'searchActions',
          description: actionbook.searchActions.description,
          parameters: actionbook.searchActions.params.json,
        },
      },
      {
        type: 'function',
        function: {
          name: 'getActionById',
          description: actionbook.getActionById.description,
          parameters: actionbook.getActionById.params.json,
        },
      },
    ]

    const completion = await openai.chat.completions.create({
      model: 'gpt-4o',
      tools,
      messages: [{ role: 'user', content: 'Search for Google login actions' }],
    })
    ```
  </Accordion>

  <Accordion title="With Anthropic Claude SDK">
    ```typescript theme={null}
    import { Actionbook } from '@actionbookdev/sdk'
    import Anthropic from '@anthropic-ai/sdk'

    const actionbook = new Actionbook()
    const anthropic = new Anthropic()

    const tools: Anthropic.Tool[] = [
      {
        name: 'searchActions',
        description: actionbook.searchActions.description,
        input_schema: actionbook.searchActions.params.json,
      },
      {
        name: 'getActionById',
        description: actionbook.getActionById.description,
        input_schema: actionbook.getActionById.params.json,
      },
    ]

    const message = await anthropic.messages.create({
      model: 'claude-sonnet-4-20250514',
      max_tokens: 1024,
      tools,
      messages: [{ role: 'user', content: 'Search for Twitter post actions' }],
    })
    ```
  </Accordion>

  <Accordion title="With Google Gemini SDK">
    ```typescript theme={null}
    import { Actionbook } from '@actionbookdev/sdk'
    import { GoogleGenAI } from '@google/genai'

    const actionbook = new Actionbook()
    const genai = new GoogleGenAI({ apiKey: process.env.GEMINI_API_KEY })

    const tools = [
      {
        functionDeclarations: [
          {
            name: 'searchActions',
            description: actionbook.searchActions.description,
            parameters: actionbook.searchActions.params.json,
          },
          {
            name: 'getActionById',
            description: actionbook.getActionById.description,
            parameters: actionbook.getActionById.params.json,
          },
        ],
      },
    ]

    const response = await genai.models.generateContent({
      model: 'gemini-2.0-flash',
      contents: [{ role: 'user', parts: [{ text: 'Search for YouTube upload actions' }] }],
      config: { tools },
    })
    ```
  </Accordion>

  <Accordion title="With Stagehand">
    ```typescript theme={null}
    import { Actionbook } from '@actionbookdev/sdk'
    import { tool } from 'ai'

    const actionbook = new Actionbook()

    const agent = stagehand.agent({
      model: 'openai/gpt-4o',
      tools: {
        searchActions: tool({
          description: actionbook.searchActions.description,
          inputSchema: actionbook.searchActions.params.zod,
          execute: async ({ query }) => actionbook.searchActions(query),
        }),
        getActionById: tool({
          description: actionbook.getActionById.description,
          inputSchema: actionbook.getActionById.params.zod,
          execute: async ({ id }) => actionbook.getActionById(id),
        }),
      },
    })

    await agent.execute('Search for Airbnb booking actions and get the action manual')
    ```
  </Accordion>
</AccordionGroup>

## Installation

```bash theme={null}
# Using npm
npm install @actionbookdev/sdk

# Using pnpm
pnpm add @actionbookdev/sdk

# Using yarn
yarn add @actionbookdev/sdk

# Using bun
bun add @actionbookdev/sdk
```

## Basic Usage

```typescript theme={null}
import { Actionbook } from '@actionbookdev/sdk'

// Initialize the client (no API key required during open beta)
const actionbook = new Actionbook()

// Search for action manuals
const results = await actionbook.searchActions('airbnb search')
console.log(`Found ${results.length} actions:`, results)

// Get a specific action by ID
const action = await actionbook.getActionById(results[0].id)
console.log('Action details:', action)

// Access the selectors
const selector =
  action.selectors.css ||
  action.selectors.dataTestId ||
  action.selectors.ariaLabel

console.log('Use this selector:', selector)
```

## Tool Definitions

Each method has `description` and `params` attached for easy integration with any LLM framework.

```typescript theme={null}
import { Actionbook } from '@actionbookdev/sdk'

const actionbook = new Actionbook()

// Description
actionbook.searchActions.description // "Search for action manuals by keyword"

// Params - JSON Schema format
actionbook.searchActions.params.json // { type: "object", properties: { query: { type: "string" } }, required: ["query"] }

// Params - Zod format
actionbook.searchActions.params.zod // z.object({ query: z.string() })
```