LiteWebAgent backend

Please note that this is a backend-only service. We have moved this backend to the monorepo: LiteWebAgent along with the frontend.

1. QuickStart

python3.11 -m venv venv
. venv/bin/activate
pip3.11 install -r requirements.txt

Then, a required step is to setup playwright by running

playwright install chromium

test your playwright installation

from playwright.sync_api import sync_playwright

with sync_playwright() as p:
    browser = p.chromium.launch(headless=True)  # Set headless=True if you prefer no UI
    page = browser.new_page()
    page.goto('https://example.com')
    print(page.title())
    browser.close()

Then please create a .env file, and update your API keys:

cp .env.example .env

You are ready to go! Try FunctionCallingAgent on google.com

python3.11 api/google_test.py

start local fastapi server

python3.11 -m api.main

2. Local testing

start browserbase session, retrieve live_browser_url for the frontend

curl -X POST 'http://0.0.0.0:8000/start-browserbase' \
-H 'Content-Type: application/json' \
-d '{"storage_state_s3_path": null}'
{"live_browser_url":"https://www.browserbase.com/devtools-fullscreen/inspector.html?wss=connect.browserbase.com/debug/f525ba67-c88e-4485-b207-dd9bf188729f/devtools/page/1467F73862C1EB9B7C68C41A4C654BD6?debug=true","session_id":"f525ba67-c88e-4485-b207-dd9bf188729f","status":"started","storage_state_path":null}%   

set goal, and some initial steps

curl -X POST 'http://0.0.0.0:8000/run-agent-initial-steps-stream' \
-H 'Content-Type: application/json' \
-d '{
  "session_id": "f525ba67-c88e-4485-b207-dd9bf188729f",
  "starting_url": "https://www.google.com",
  "goal": "type dining table in text box",
  "s3_path": "s3://test-litewebagent/adhoc-test/flow.json",
  "storage_state_s3_path": null
}'



data: {"type": "status", "message": "Starting setup..."}

data: {"type": "status", "message": "Storage state loaded"}

data: {"type": "browser", "message": "Browser connected at https://www.browserbase.com/devtools-fullscreen/inspector.html?wss=connect.browserbase.com/debug/f525ba67-c88e-4485-b207-dd9bf188729f/devtools/page/1467F73862C1EB9B7C68C41A4C654BD6?debug=true"}

data: {"type": "status", "message": "Agent setup complete"}

data: {"type": "thinking", "message": "Processing next step..."}

data: {"type": "tool_calls", "message": "Executing 1 actions..."}

data: {"type": "tool_execution", "message": "Executing: navigation"}

data: {"type": "tool_result", "message": {"tool_call_id": "call_UAJ8YtGfkMmIKE3QCZ5XgFnn", "role": "tool", "name": "navigation", "content": "The action is: ```fill('90', 'dining table')``` - the result is: Yes, the goal is finished. The screenshot shows that the text \"dining table\" has been typed into the Google search box, as evidenced by its presence in the text input field. This aligns with the original goal of typing \"dining table\" in the text box."}}

data: {"type": "thinking", "message": "The task of typing \"dining table\" in the text box has been completed. Please provide further instructions if needed."}

data: {"type": "action", "message": "No actions needed"}

data: {"type": "complete", "message": "Task completed", "response": [{"finish_reason": "stop", "index": 0, "message": {"content": "The task of typing \"dining table\" in the text box has been completed. Please provide further instructions if needed.", "role": "assistant", "tool_calls": null, "function_call": null}}]}

add some follow-up steps

curl -X POST 'http://0.0.0.0:8000/run-agent-followup-steps-stream' \
-H 'Content-Type: application/json' \
-d '{
  "session_id": "f525ba67-c88e-4485-b207-dd9bf188729f",
  "goal": "click on search button",
  "s3_path": "s3://test-litewebagent/adhoc-test/flow.json",
  "storage_state_s3_path": null
}'

terminate the browserbase session, whole session completed

curl -X POST "http://0.0.0.0:8000/terminate-browserbase?session_id=f525ba67-c88e-4485-b207-dd9bf188729f"
{"status":"terminated"}%

3. Production

curl -X POST 'https://lite-web-agent-backend.vercel.app/start-browserbase' \
-H 'Content-Type: application/json' \
-d '{"storage_state_s3_path": null}'  
{"live_browser_url":"https://www.browserbase.com/devtools-fullscreen/inspector.html?wss=connect.browserbase.com/debug/3744d2f6-4ba5-4d98-a44c-5b03fc7b33a3/devtools/page/30A3CDBECAE2E9B96FCE5A6FAD83D6BA?debug=true","session_id":"3744d2f6-4ba5-4d98-a44c-5b03fc7b33a3","status":"started","storage_state_path":null}

curl -X POST 'https://lite-web-agent-backend.vercel.app/run-agent-initial-steps-stream' \
-H 'Content-Type: application/json' \
-d '{
  "session_id": "3744d2f6-4ba5-4d98-a44c-5b03fc7b33a3",
  "starting_url": "https://www.google.com",
  "goal": "type dining table in text box",
  "s3_path": "s3://test-litewebagent/adhoc-test/flow.json",
  "storage_state_s3_path": null
}'

curl -X POST "https://lite-web-agent-backend.vercel.app/terminate-browserbase?session_id=3744d2f6-4ba5-4d98-a44c-5b03fc7b33a3"
{"status":"terminated"}%