Here are three different patterns for using the Simplex SDK, demonstrated through a flight search example.

The task is to find a flight from New York to San Francisco on January 1st, 2025.

​
Open Feedback Loop

The simplest implementation - continuously gets the next step and executes actions until the task is complete.

def open_feedback_loop():
    high_level_task = "Find a flight from New York to San Francisco on January 1st, 2025"
    url = "https://www.google.com/travel/flights"

    client = Simplex(api_key=os.getenv("SIMPLEX_API_KEY"), record_video=True)
    client.browser_actor.navigate_to_url(url)
    state = client.browser_actor.take_screenshot()

    while not client.is_task_complete(high_level_task, state):
        next_step = client.get_next_step(high_level_task, state)
        print(next_step)
        actions = client.step_to_action(next_step, state)
        print(actions)

        for action in actions:
            client.browser_actor.execute(action)
            state = client.browser_actor.take_screenshot()

    client.browser_actor.close()
    client.browser_actor.save_recording("output.mp4")

​
Closed Feedback Loop

Here we’ve added a critic at the end of the step. If a step is not successful, we’ll recompute and try again.

def closed_feedback_loop():
    high_level_task = "Find a flight from New York to San Francisco on January 1st, 2025"
    url = "https://www.google.com/travel/flights"

    client = Simplex(api_key=os.getenv("SIMPLEX_API_KEY"), record_video=True)
    client.browser_actor.navigate_to_url(url)
    state = client.browser_actor.take_screenshot()

    while not client.is_task_complete(high_level_task, state):

        step_complete = False

        ## add critic here for every step
        while not step_complete:
            next_step = client.get_next_step(high_level_task, state)
            actions = client.step_to_action(next_step, state)

            for action in actions:
                client.browser_actor.execute(action)
                state = client.browser_actor.take_screenshot()

            step_complete = client.is_step_complete(next_step, state)

    client.browser_actor.close()
    client.browser_actor.save_recording("output.mp4")

​
Precomputed Trajectory

You can also use your own planner or precompute a trajectory yourself. The browser actor uses your precomputed trajectory and only uses our API endpoints to find element locations and execute actions.

def precomputed_trajectory():
    client = Simplex(api_key=os.getenv("SIMPLEX_API_KEY"), record_video=True)
    high_level_task = "Find a flight from New York to San Francisco on January 1st, 2025"
    url = "https://www.google.com/travel/flights"

    client.browser_actor.navigate_to_url(url)
    state = client.browser_actor.take_screenshot()

    ## use your own planner or calculate the trajectory yourself
    trajectory = ["click on round trip", 
                  "click on one way", 
                  "type in departure of New York", 
                  "type in destination of San Francisco", 
                  "enter date of January 1st, 2025", 
                  "click on search"]

    ## can add some max iters here as well
    for next_step in trajectory:
        actions = client.step_to_action(next_step, state)

        for action in actions:
            client.browser_actor.execute(action)
            state = client.browser_actor.take_screenshot()
        
    client.browser_actor.close()
    client.browser_actor.save_recording("output.mp4")
Authentication