Notice: The Dendrite SDK is not under active development anymore. However, the project will remain fully open source so that you and others can learn from it. Feel free to fork, study, or adapt this code for your own projects as you wish – reach out to us on Discord if you have questions! We love chatting about web AI agents. 🤖

What is Dendrite?

Dendrite is a framework that makes it easy for web AI agents to browse the internet just like humans do. Use Dendrite to:

• 👆🏼 Interact with elements
• 💿 Extract structured data
• 🔓 Authenticate on websites
• ↕️ Download/upload files
• 🚫 Browse without getting blocked

A simple outlook integration

With Dendrite it's easy to create web interaction tools for your agent.

Here's how you can send an email:

from dendrite import AsyncDendrite


async def send_email(to, subject, message):
    client = AsyncDendrite(auth="outlook.live.com")

    # Navigate
    await client.goto(
        "https://outlook.live.com/mail/0/", expected_page="An email inbox"
    )

    # Create new email and populate fields
    await client.click("The new email button")
    await client.fill("The recipient field", to)
    await client.press("Enter")
    await client.fill("The subject field", subject)
    await client.fill("The message field", message)

    # Send email
    await client.press("Enter", hold_cmd=True)


if __name__ == "__main__":
    import asyncio

    asyncio.run(send_email("[email protected]", "Hello", "This is a test email"))

You'll need to add your own Anthropic key or configure which LLMs to use yourself.

ANTHROPIC_API_KEY=sk-...

To authenticate on any web service with Dendrite, follow these steps:

1. Run the authentication command

dendrite auth --url outlook.live.com

2. This command will open a browser that you'll be able to login with.

3. After you've logged in, press enter in your terminal. This will save your cookies locally so that they can be used in your code.

Read more about authentication in our docs.

Quickstart

pip install dendrite && dendrite install

Simple navigation and interaction

from dendrite import AsyncDendrite

async def main():
    client = AsyncDendrite()

    await client.goto("https://google.com")
    await client.fill("Search field", "Hello world")
    await client.press("Enter")

if __name__ == "__main__":
    import asyncio
    asyncio.run(main())

In the example above, we simply go to Google, populate the search field with "Hello world" and simulate a keypress on Enter. It's a simple example that starts to explore the endless possibilities with Dendrite. Now you can create tools for your agents that have access to the full web without depending on APIs.

More Examples

Get any page as markdown

This is a simple example of how to get any page as markdown, great for feeding to an LLM.

from dendrite import AsyncDendrite
from dotenv import load_dotenv

async def main():
    browser = AsyncDendrite()

    await browser.goto("https://dendrite.systems")
    await browser.wait_for("the page to load")

    # Get the entire page as markdown
    md = await browser.markdown()
    print(md)
    print("=" * 200)

    # Only get a certain part of the page as markdown
    data_extraction_md = await browser.markdown("the part about data extraction")
    print(data_extraction_md)

if __name__ == "__main__":
    import asyncio
    asyncio.run(main())

Get Company Data from Y Combinator

The classic web data extraction test, made easy:

from dendrite import AsyncDendrite
import pprint
import asyncio


async def main():
    browser = AsyncDendrite()

    # Navigate
    await browser.goto("https://www.ycombinator.com/companies")

    # Find and fill the search field with "AI agent"
    await browser.fill(
        "Search field", value="AI agent"
    )  # Element selector cached since before
    await browser.press("Enter")

    # Extract startups with natural language description
    # Once created by our agent, the same script will be cached and reused
    startups = await browser.extract(
        "All companies. Return a list of dicts with name, location, description and url"
    )
    pprint.pprint(startups, indent=2)


if __name__ == "__main__":
    asyncio.run(main())

returns

[ { 'description': 'Book accommodations around the world.',
    'location': 'San Francisco, CA, USA',
    'name': 'Airbnb',
    'url': 'https://www.ycombinator.com/companies/airbnb'},
  { 'description': 'Digital Analytics Platform',
    'location': 'San Francisco, CA, USA',
    'name': 'Amplitude',
    'url': 'https://www.ycombinator.com/companies/amplitude'},
...
] }

Extract Data from Google Analytics

Here's how to get the amount of monthly visitors from Google Analytics using the extract function:

async def get_visitor_count() -> int:
    client = AsyncDendrite(auth="analytics.google.com")

    await client.goto(
      "https://analytics.google.com/analytics/web",
      expected_page="Google Analytics dashboard"
    )

    # The Dendrite extract agent will create a web script that is cached
    # and reused. It will self-heal when the website updates
    visitor_count = await client.extract("The amount of visitors this month", int)
    return visitor_count

Download Bank Transactions

Here's tool that allows our AI agent to download our bank's monthly transactions so that they can be analyzed and compiled into a report.

from dendrite import AsyncDendrite

async def get_transactions() -> str:
    client = AsyncDendrite(auth="mercury.com")

    # Navigate and wait for loading
    await client.goto(
      "https://app.mercury.com/transactions",
      expected_page="Dashboard with transactions"
    )
    await client.wait_for("The transactions to finish loading")

    # Modify filters
    await client.click("The 'add filter' button")
    await client.click("The 'show transactions for' dropdown")
    await client.click("The 'this month' option")

    # Download file
    await client.click("The 'export filtered' button")
    transactions = await client.get_download()

    # Save file locally
    path = "files/transactions.xlsx"
    await transactions.save_as(path)

    return path

async def analyze_transactions(path: str):
    ... # Analyze the transactions with LLM of our choice

Documentation

Read the full docs here

Find more advanced examples in our docs.

Remote Browsers

When you want to scale up your AI agents, we support using browsers hosted by Browserbase. This way you can run many agents in parallel without having to worry about the infrastructure.

To start using Browserbase just swap out the AsyncDendrite class with AsyncDendriteRemoteBrowser and add your Browserbase API key and project id, either in the code or in a .env file like this:

# ... previous keys
BROWSERBASE_API_KEY=
BROWSERBASE_PROJECT_ID=

# from dendrite import AsyncDendrite
from dendrite import AsyncDendriteRemoteBrowser

async def main():
    # client = AsyncDendrite(...)
    client = AsyncDendriteRemoteBrowser(
        # Use interchangeably with the AsyncDendrite class
        browserbase_api_key="...", # or specify the browsebase keys in the .env file
        browserbase_project_id="..."
    )
    ...

if __name__ == "__main__":
    import asyncio
    asyncio.run(main())