5 tips to automate the browser with browser-use 1.0

The repository is available at github.com/browser-use/browser-use, under MIT license, and is actively maintained by the browser-use team. Version 1.0 was published on November 10, 2026.

In this article you will see:

• What changed in version 1.0
• How to install and run your first agent
• How to use ultrafast mode for repetitive tasks
• How to keep the session alive between runs
• How to integrate with external language models

What changed in version 1.0

Version 1.0 brings a restructuring centered on CDP. According to the release notes, the 0.x model gateway (bu-2-0, bu-1-0, bu-latest) was discontinued. The environment variable BROWSER_USE_LLM_URL is no longer read: anyone who needs a custom endpoint should use ChatBrowserUse with base_url. LangChain models now raise TypeError instead of being silently accepted.

Another important change is the keep_alive behavior. A run after a dead browser daemon continues in the agent tab, instead of falling back to about:blank.

Version 1.0 also includes improvements to page observation, form state, file access and browser recovery, plus support for Browser Use Cloud SDK V4.

Tip 1: install and run your first agent

Installation is done via pip or uv. The project requires Python 3.11 or higher.

$ pip install browser-use

Or with uv:

$ uv add browser-use

To run your first agent, set the model API key and run the browser-use command with a task. The project provides a CLI with five entries. The documentation includes an updated quickstart for version 1.0.

Tip 2: use ultrafast mode for repetitive tasks

Version 1.0 introduces ultrafast mode, used internally by bu-fast. According to the benchmark published in the release notes, ultrafast mode reached a score of 62.5 on BU Bench v2 (60 tasks, 3 repetitions, cloud browsers, November 9, 2026), with a cost of $0.070 per task and 650 seconds per task.

This mode is useful for repetitive tasks at scale, such as collecting data from pages or filling out forms. The mode prioritizes speed over deep reasoning, which makes it suitable for subtasks that a larger agent delegates.

Tip 3: keep the session alive between runs

The keep_alive behavior in version 1.0 allows a run to continue in the agent tab even after the browser daemon has died. This avoids losing page context between calls.

For those running agents in pipelines, this means the page state can be preserved between steps, reducing page reopening time and context reload.

Tip 4: integrate with external language models

browser-use supports integration with external language models. Version 1.0 removes silent support for LangChain models (now raises TypeError), which forces explicit configuration.

To use a custom model, pass a ChatBrowserUse instance with the base_url parameter pointing to the desired endpoint. The environment variable BROWSER_USE_LLM_URL is no longer read.

The project also offers integration with the OpenAI API and supports the MCP protocol, allowing external agents to use the browser as a tool.