Deterministic rerun

Have the agent save and test a reusable script, then run it again from the same workspace.

For repeated workflows, create a dedicated workspace and ask the agent to turn its successful process into a script. The important part is explicit: tell it to reproduce what it just did, test the script, and save instructions for the next run.

You can create the workspace in the dashboard or through the API:

Python

from browser_use_sdk.v4 import AsyncBrowserUse

client = AsyncBrowserUse()
workspace = await client.workspaces.create(name="hn-scraper")

created = await client.runs.create(
    """
    Get the top five Hacker News stories as JSON.
    Then create helper functions or a script that performs exactly what you did.
    Test it, save it as scripts/hn_top.py, and save reuse instructions in
    scripts/README.md.
    """,
    workspace_id=workspace.id,
)
first = await client.runs.wait_for_completion(created.id)
print(first.result)

Later, start a new run in the same workspace and tell the agent to use the saved script:

Python

created = await client.runs.create(
    "Run scripts/hn_top.py for the top 10 stories. Use the existing script; fix and retest it only if needed.",
    workspace_id=workspace.id,
)
rerun = await client.runs.wait_for_completion(created.id)
print(rerun.result)

This pattern gives the agent a fast, inspectable path and lets it repair the script when the website changes. Keep one workspace per workflow so scripts, fixtures, outputs, and instructions stay together.

V4 does not automatically turn a task into a cached $0-LLM execution. Each rerun starts an agent, so it still has token cost. The saved script usually makes the run faster and cheaper, but you should measure it for your workflow.