Open-source Python library for AI browser agents, plus Browser Use Cloud: stealth browsers and a hosted agent API.
Python ai-agentsai-toolsbrowser-agentbrowser-automationbrowser-usecdpcloud-browsercomputer-useheadless-browserllmmcpplaywrightpythonremote-browserstealth-browserweb-automationweb-scraping
remove `run` command and agent infrastructure from CLI
ShawnPana0dc2fc5a6fb7
diff --git a/browser_use/skill_cli/README.md b/browser_use/skill_cli/README.mdindex f491c8e75..7efed3f77 100644--- a/browser_use/skill_cli/README.md+++ b/browser_use/skill_cli/README.md@@ -190,20 +190,6 @@ browser-use python --reset # Clear namespacebrowser-use python --file script.py # Run Python file```-## Agent Tasks--Run AI-powered browser automation tasks.--### Local Mode-```bash-browser-use run "Fill the contact form with test data"-browser-use run "Extract all product prices" --max-steps 50-browser-use run "task" --llm gpt-4o # Specify LLM model-```--Requires an LLM API key (`OPENAI_API_KEY`, `ANTHROPIC_API_KEY`, etc.).--## Cloud APIGeneric REST passthrough to the Browser-Use Cloud API.@@ -277,7 +263,6 @@ browser-use open https://abc.trycloudflare.com| `--headed` | Show browser window || `--profile [NAME]` | Use real Chrome (bare `--profile` uses "Default") || `--json` | Output as JSON |-| `--api-key KEY` | Override API key || `--mcp` | Run as MCP server via stdin/stdout |## Examplesdiff --git a/browser_use/skill_cli/__init__.py b/browser_use/skill_cli/__init__.pyindex 72d560e0d..dd6f24c90 100644--- a/browser_use/skill_cli/__init__.py+++ b/browser_use/skill_cli/__init__.py@@ -8,7 +8,6 @@ Usage:browser-use click 5browser-use type "Hello World"browser-use python "print(browser.url)"- browser-use run "Fill the contact form"browser-use close"""diff --git a/browser_use/skill_cli/commands/__init__.py b/browser_use/skill_cli/commands/__init__.pyindex a94353af7..37e4849ad 100644--- a/browser_use/skill_cli/commands/__init__.py+++ b/browser_use/skill_cli/commands/__init__.py@@ -1,7 +1,6 @@"""Command handlers for browser-use CLI."""from browser_use.skill_cli.commands import (- agent,browser,doctor,python_exec,@@ -9,7 +8,6 @@ from browser_use.skill_cli.commands import ()__all__ = [- 'agent','browser','doctor','python_exec',diff --git a/browser_use/skill_cli/commands/agent.py b/browser_use/skill_cli/commands/agent.pydeleted file mode 100644index 7325995a7..000000000--- a/browser_use/skill_cli/commands/agent.py+++ /dev/null@@ -1,159 +0,0 @@-"""Agent task command handler."""--import logging-import os-from typing import Any--from browser_use.skill_cli.sessions import SessionInfo--logger = logging.getLogger(__name__)---async def handle(session: SessionInfo, params: dict[str, Any]) -> Any:- """Handle agent run command.-- Runs a task using the local browser-use agent.- """- task = params.get('task')- if not task:- return {'success': False, 'error': 'No task provided'}-- return await _handle_local_task(session, params)---async def _handle_local_task(session: SessionInfo, params: dict[str, Any]) -> Any:- """Handle task execution locally with browser-use agent."""- task = params['task']- max_steps = params.get('max_steps')- model = params.get('llm') # Optional model override-- try:- # Import agent and LLM- from browser_use.agent.service import Agent-- # Try to get LLM from environment (with optional model override)- llm = await get_llm(model=model)- if llm is None:- if model:- return {- 'success': False,- 'error': f'Could not initialize model "{model}". '- f'Make sure the appropriate API key is set (OPENAI_API_KEY, ANTHROPIC_API_KEY, or GOOGLE_API_KEY).',- }- return {- 'success': False,- 'error': 'No LLM configured. Set OPENAI_API_KEY, ANTHROPIC_API_KEY, or GOOGLE_API_KEY',- }-- # Create and run agent- agent = Agent(- task=task,- llm=llm,- browser_session=session.browser_session,- )-- logger.info(f'Running local agent task: {task}')- run_kwargs = {}- if max_steps is not None:- run_kwargs['max_steps'] = max_steps- result = await agent.run(**run_kwargs)-- # Extract result info- final_result = result.final_result() if result else None-- return {- 'success': True,- 'task': task,- 'steps': len(result) if result else 0,- 'result': str(final_result) if final_result else None,- 'done': result.is_done() if result else False,- }-- except Exception as e:- logger.exception(f'Local agent task failed: {e}')- return {- 'success': False,- 'error': str(e),- 'task': task,- }---def _get_verified_models() -> dict[str, set[str]]:- """Extract verified model names from SDK sources of truth."""- import typing-- from anthropic.types.model_param import ModelParam- from openai.types.shared.chat_model import ChatModel-- from browser_use.llm.google.chat import VerifiedGeminiModels-- # OpenAI: ChatModel is a Literal type- openai_models = set(typing.get_args(ChatModel))-- # Anthropic: ModelParam is Union[Literal[...], str] - extract the Literal- anthropic_literal = typing.get_args(ModelParam)[0]- anthropic_models = set(typing.get_args(anthropic_literal))-- # Google: VerifiedGeminiModels Literal- google_models = set(typing.get_args(VerifiedGeminiModels))-- return {- 'openai': openai_models,- 'anthropic': anthropic_models,- 'google': google_models,- }---_VERIFIED_MODELS: dict[str, set[str]] | None = None---def _get_provider_for_model(model: str) -> str | None:- """Determine the provider by checking SDK verified model lists."""- global _VERIFIED_MODELS- if _VERIFIED_MODELS is None:- _VERIFIED_MODELS = _get_verified_models()-- for provider, models in _VERIFIED_MODELS.items():- if model in models:- return provider-- return None---def get_llm(model: str | None = None) -> Any:- """Get LLM instance from environment configuration.-- Args:- model: Optional model name to use. If provided, will instantiate- the appropriate provider for that model. If not provided,- auto-detects from available API keys.-- Supported providers: OpenAI, Anthropic, Google.- Model names are validated against each SDK's verified model list.- """- from browser_use.llm import ChatAnthropic, ChatGoogle, ChatOpenAI-- if model:- provider = _get_provider_for_model(model)-- if provider == 'openai':- return ChatOpenAI(model=model)- elif provider == 'anthropic':- return ChatAnthropic(model=model)- elif provider == 'google':- return ChatGoogle(model=model)- else:- logger.warning(f'Unknown model: {model}. Not in any verified model list.')- return None-- # No model specified - auto-detect from available API keys- if os.environ.get('OPENAI_API_KEY'):- return ChatOpenAI(model='o3')-- if os.environ.get('ANTHROPIC_API_KEY'):- return ChatAnthropic(model='claude-sonnet-4-0')-- if os.environ.get('GOOGLE_API_KEY'):- return ChatGoogle(model='gemini-flash-latest')-- return Nonediff --git a/browser_use/skill_cli/commands/browser.py b/browser_use/skill_cli/commands/browser.pyindex 658808298..34a97d077 100644--- a/browser_use/skill_cli/commands/browser.py+++ b/browser_use/skill_cli/commands/browser.py@@ -246,7 +246,7 @@ async def handle(action: str, session: SessionInfo, params: dict[str, Any]) -> Aquery = params['query']# This requires LLM integration# For now, return a placeholder- return {'query': query, 'error': 'extract requires agent mode - use: browser-use run "extract ..."'}+ return {'query': query, 'error': 'extract is not yet implemented'}elif action == 'hover':index = params['index']diff --git a/browser_use/skill_cli/daemon.py b/browser_use/skill_cli/daemon.pyindex 7f40190bf..9fa33d5f5 100644--- a/browser_use/skill_cli/daemon.py+++ b/browser_use/skill_cli/daemon.py@@ -137,7 +137,7 @@ class Daemon:},}- from browser_use.skill_cli.commands import agent, browser, python_exec+ from browser_use.skill_cli.commands import browser, python_exec# Get or create the single sessionsession = await self._get_or_create_session()@@ -147,8 +147,6 @@ class Daemon:result = await browser.handle(action, session, params)elif action == 'python':result = await python_exec.handle(session, params)- elif action == 'run':- result = await agent.handle(session, params)else:return {'id': req_id, 'success': False, 'error': f'Unknown action: {action}'}diff --git a/browser_use/skill_cli/main.py b/browser_use/skill_cli/main.pyindex fdbc487fe..34e1c7f89 100755--- a/browser_use/skill_cli/main.py+++ b/browser_use/skill_cli/main.py@@ -177,7 +177,6 @@ def _is_daemon_alive() -> bool:def ensure_daemon(headed: bool,profile: str | None,- api_key: str | None,*,explicit_config: bool = False,) -> None:@@ -213,8 +212,6 @@ def ensure_daemon(# Set up environmentenv = os.environ.copy()- if api_key:- env['BROWSER_USE_API_KEY'] = api_key# Start daemon as background processif sys.platform == 'win32':@@ -283,13 +280,7 @@ def build_parser() -> argparse.ArgumentParser:# Build epilogepilog_parts = []- epilog_parts.append("""Local Mode (default):- browser-use run "Fill the form" # Uses local browser + your API keys- browser-use run "task" --llm gpt-4o # Specify model (requires API key)- browser-use open https://example.com""")-- epilog_parts.append("""-Cloud API:+ epilog_parts.append("""Cloud API:browser-use cloud login <api-key> # Save API keybrowser-use cloud v2 GET /browsers # List browsersbrowser-use cloud v2 POST /tasks '{...}' # Create task@@ -298,6 +289,7 @@ Cloud API:epilog_parts.append("""Setup:+ browser-use open https://example.com # Navigate to URLbrowser-use install # Install Chromium browserbrowser-use init # Generate template file""")@@ -318,7 +310,6 @@ Setup:help='Use real Chrome with profile (bare --profile uses "Default")',)parser.add_argument('--json', action='store_true', help='Output as JSON')- parser.add_argument('--api-key', help='LLM API key')parser.add_argument('--mcp', action='store_true', help='Run as MCP server (JSON-RPC via stdin/stdout)')parser.add_argument('--template', help='Generate template file (use with --output for custom path)')@@ -512,16 +503,6 @@ Setup:p.add_argument('--reset', action='store_true', help='Reset Python namespace')p.add_argument('--vars', action='store_true', help='Show defined variables')- # -------------------------------------------------------------------------- # Agent Tasks- # --------------------------------------------------------------------------- p = subparsers.add_parser('run', help='Run agent task (requires API key)')- p.add_argument('task', help='Task description')- p.add_argument('--max-steps', type=int, help='Maximum steps')- # Model selection- p.add_argument('--llm', help='LLM model (gpt-4o, claude-sonnet-4-20250514, gemini-2.0-flash)')-# -------------------------------------------------------------------------# Tunnel Commands# -------------------------------------------------------------------------@@ -732,18 +713,14 @@ def main() -> int:print('No active browser session')return 0- # Set API key in environment if provided- if args.api_key:- os.environ['BROWSER_USE_API_KEY'] = args.api_key-# Ensure daemon is running# Only restart on config mismatch if the user explicitly passed config flagsexplicit_config = any(flag in sys.argv for flag in ('--headed', '--profile'))- ensure_daemon(args.headed, args.profile, args.api_key, explicit_config=explicit_config)+ ensure_daemon(args.headed, args.profile, explicit_config=explicit_config)# Build params from argsparams = {}- skip_keys = {'command', 'headed', 'json', 'api_key'}+ skip_keys = {'command', 'headed', 'json'}for key, value in vars(args).items():if key not in skip_keys and value is not None:diff --git a/skills/browser-use/SKILL.md b/skills/browser-use/SKILL.mdindex 7cbba789c..a6257a6fe 100644--- a/skills/browser-use/SKILL.md+++ b/skills/browser-use/SKILL.md@@ -69,10 +69,6 @@ browser-use wait text "Success" # Wait for text# Sessionbrowser-use close # Close browser session--# AI Agent-browser-use run "Fill the contact form" # Run local agent-browser-use run "task" --llm gpt-4o # Specify model```## Commands@@ -165,14 +161,6 @@ The Python session maintains state across commands. The `browser` object provide- `browser.screenshot(path)`, `browser.scroll(direction, amount)` — visual- `browser.wait(seconds)`, `browser.extract(query)` — utilities-### Agent Tasks-```bash-browser-use run "Fill the contact form with test data" # AI agent-browser-use run "Extract all product prices" --max-steps 50-browser-use run "task" --llm gpt-4o # Specify LLM model-browser-use run "task" --llm claude-sonnet-4-20250514-```-### Cloud API```bashbrowser-use cloud login <api-key> # Save API keydiff --git a/skills/remote-browser/SKILL.md b/skills/remote-browser/SKILL.mdindex 58b7e0268..80fbb55bb 100644--- a/skills/remote-browser/SKILL.md+++ b/skills/remote-browser/SKILL.md@@ -67,10 +67,6 @@ browser-use wait text "Success" # Wait for text# Sessionbrowser-use close # Close browser session--# AI Agent-browser-use run "Fill the contact form" # Run local agent-browser-use run "task" --llm gpt-4o # Specify model```## Commands@@ -161,14 +157,6 @@ The Python session maintains state across commands. The `browser` object provide- `browser.screenshot(path)`, `browser.scroll(direction, amount)` — visual- `browser.wait(seconds)`, `browser.extract(query)` — utilities-### Agent Tasks-```bash-browser-use run "Fill the contact form with test data" # AI agent-browser-use run "Extract all product prices" --max-steps 50-browser-use run "task" --llm gpt-4o # Specify LLM model-browser-use run "task" --llm claude-sonnet-4-20250514-```-### Tunnels```bashbrowser-use tunnel <port> # Start tunnel (returns URL)