browser-use/browser-use · error · AttributeError

module '{__name__}' has no attribute '{name}'

Error message

module '{__name__}' has no attribute '{name}'

What it means

The dynamic extraction-schema compiler (used by the `extract` tool to turn a JSON Schema into a Pydantic model) only supports a structural subset of JSON Schema. `_check_unsupported` rejects composition/reference keywords listed in _UNSUPPORTED_KEYWORDS ($ref, allOf, anyOf, oneOf, not, etc.) because there is no reliable translation to a flat create_model field map. The error is raised at schema-compile time, before any LLM call or page extraction happens.

Source

Thrown at browser_use/__init__.py:135

	"""Lazy import mechanism - only import modules when they're actually accessed."""
	if name in _LAZY_IMPORTS:
		module_path, attr_name = _LAZY_IMPORTS[name]
		try:
			from importlib import import_module

			module = import_module(module_path)
			if attr_name is None:
				# For modules like 'models', return the module itself
				attr = module
			else:
				attr = getattr(module, attr_name)
			# Cache the imported attribute in the module's globals
			globals()[name] = attr
			return attr
		except ImportError as e:
			raise ImportError(f'Failed to import {name} from {module_path}: {e}') from e

	raise AttributeError(f"module '{__name__}' has no attribute '{name}'")


__all__ = [
	'Agent',
	'BrowserSession',
	'Browser',  # Alias for BrowserSession
	'BrowserProfile',
	'Controller',
	'DomService',
	'SystemPrompt',
	'ActionResult',
	'ActionModel',
	'AgentHistoryList',
	# Chat models
	'ChatOpenAI',
	'ChatGoogle',
	'ChatAnthropic',
	'ChatAnthropicBedrock',

View on GitHub (pinned to 6c73fced2f)

Solutions

1. Flatten the schema by hand: inline every $ref, and replace anyOf/oneOf/allOf unions with a single concrete type or a type array like {"type": ["string", "null"]} if supported by your version.
2. Replace optional-union patterns with the property flagged in `required` omission plus a nullable type representation the compiler accepts.
3. Keep extraction schemas small and hand-written for this tool — a flat object of scalar/array/object properties with enums as string enums.
4. If you need full JSON Schema unions, define a static Pydantic model and pass that instead of a raw schema where the API allows it.

Example fix

# before
schema = {
  "type": "object",
  "properties": {
    "price": {"anyOf": [{"type": "number"}, {"type": "null"}]},
    "address": {"$ref": "#/$defs/address"}
  }
}

# after
schema = {
  "type": "object",
  "properties": {
    "price": {"type": "number"},
    "address": {"type": "object", "properties": {"city": {"type": "string"}, "zip": {"type": "string"}}}
  }
}

UNSUPPORTED = {'$ref', '$defs', 'allOf', 'anyOf', 'oneOf', 'not'}

def schema_is_supported(node: dict) -> bool:
    if not isinstance(node, dict):
        return True
    if any(kw in node for kw in UNSUPPORTED):
        return False
    return all(schema_is_supported(v) for v in node.get('properties', {}).values())

def is_flat_extraction_schema(schema: dict) -> bool:
    return (
        isinstance(schema, dict)
        and schema.get('type') == 'object'
        and bool(schema.get('properties'))
        and schema_is_supported(schema)
    )

from browser_use.tools.extraction.schema_utils import schema_to_model
try:
    model = schema_to_model(schema)
except ValueError as e:
    if 'Unsupported JSON Schema keyword' in str(e):
        schema = flatten_refs_and_unions(schema)  # your normalizer
        model = schema_to_model(schema)

When it happens

Trigger: Passing an `output_model_schema` / extraction schema that contains any of the unsupported keywords at the top level or in any nested node (e.g. {"type":"object","properties":{"x":{"anyOf":[{"type":"string"},{"type":"null"}]}}} or a $ref to a shared definition). Also triggered by OpenAPI-derived schemas that ubiquitously use $ref/anyOf.

Common situations: Copying a schema from an OpenAPI spec or JSON-Schema-2020-12 source into extract(); modeling optional fields with anyOf [T, null] instead of type arrays; trying to reuse definitions via $defs/$ref for DRY schemas.

Related errors

• Failed to find DOM element based on backendNodeId, maybe pag
• Failed to click element: {js_e}
• Failed to import {name} from {module_path}: {e}
• Failed to click element: {e}
• Failed to get object ID for element

AI-assisted analysis of browser-use/browser-use@6c73fced2f (2026-08-14). Data as JSON: /api/errors/5472ca6d75806be8. Report an issue: GitHub.