Skip to content

Open-source Python library for AI browser agents, plus Browser Use Cloud: stealth browsers and a hosted agent API.

kv caching for BU agents

mertunsalldc0db88215ad
diff --git a/browser_use/agent/service.py b/browser_use/agent/service.py
index 1b2bc181b..05be0364d 100644
--- a/browser_use/agent/service.py
+++ b/browser_use/agent/service.py
@@ -1599,7 +1599,7 @@ class Agent(Generic[Context, AgentStructuredOutput]):
# Build kwargs for ainvoke
# Note: ChatBrowserUse will automatically generate action descriptions from output_format schema
- kwargs: dict = {'output_format': self.AgentOutput}
+ kwargs: dict = {'output_format': self.AgentOutput, 'session_id': self.session_id}
try:
response = await self.llm.ainvoke(input_messages, **kwargs)
diff --git a/browser_use/llm/anthropic/chat.py b/browser_use/llm/anthropic/chat.py
index 076d5b672..3e4afebc4 100644
--- a/browser_use/llm/anthropic/chat.py
+++ b/browser_use/llm/anthropic/chat.py
@@ -127,13 +127,15 @@ class ChatAnthropic(BaseChatModel):
return usage
@overload
- async def ainvoke(self, messages: list[BaseMessage], output_format: None = None) -> ChatInvokeCompletion[str]: ...
+ async def ainvoke(
+ self, messages: list[BaseMessage], output_format: None = None, **kwargs: Any
+ ) -> ChatInvokeCompletion[str]: ...
@overload
- async def ainvoke(self, messages: list[BaseMessage], output_format: type[T]) -> ChatInvokeCompletion[T]: ...
+ async def ainvoke(self, messages: list[BaseMessage], output_format: type[T], **kwargs: Any) -> ChatInvokeCompletion[T]: ...
async def ainvoke(
- self, messages: list[BaseMessage], output_format: type[T] | None = None
+ self, messages: list[BaseMessage], output_format: type[T] | None = None, **kwargs: Any
) -> ChatInvokeCompletion[T] | ChatInvokeCompletion[str]:
anthropic_messages, system_prompt = AnthropicMessageSerializer.serialize_messages(messages)
diff --git a/browser_use/llm/aws/chat_anthropic.py b/browser_use/llm/aws/chat_anthropic.py
index 233ebf047..8515f5660 100644
--- a/browser_use/llm/aws/chat_anthropic.py
+++ b/browser_use/llm/aws/chat_anthropic.py
@@ -147,13 +147,15 @@ class ChatAnthropicBedrock(ChatAWSBedrock):
return usage
@overload
- async def ainvoke(self, messages: list[BaseMessage], output_format: None = None) -> ChatInvokeCompletion[str]: ...
+ async def ainvoke(
+ self, messages: list[BaseMessage], output_format: None = None, **kwargs: Any
+ ) -> ChatInvokeCompletion[str]: ...
@overload
- async def ainvoke(self, messages: list[BaseMessage], output_format: type[T]) -> ChatInvokeCompletion[T]: ...
+ async def ainvoke(self, messages: list[BaseMessage], output_format: type[T], **kwargs: Any) -> ChatInvokeCompletion[T]: ...
async def ainvoke(
- self, messages: list[BaseMessage], output_format: type[T] | None = None
+ self, messages: list[BaseMessage], output_format: type[T] | None = None, **kwargs: Any
) -> ChatInvokeCompletion[T] | ChatInvokeCompletion[str]:
anthropic_messages, system_prompt = AnthropicMessageSerializer.serialize_messages(messages)
diff --git a/browser_use/llm/aws/chat_bedrock.py b/browser_use/llm/aws/chat_bedrock.py
index 39bf8cd70..610379f90 100644
--- a/browser_use/llm/aws/chat_bedrock.py
+++ b/browser_use/llm/aws/chat_bedrock.py
@@ -157,13 +157,15 @@ class ChatAWSBedrock(BaseChatModel):
)
@overload
- async def ainvoke(self, messages: list[BaseMessage], output_format: None = None) -> ChatInvokeCompletion[str]: ...
+ async def ainvoke(
+ self, messages: list[BaseMessage], output_format: None = None, **kwargs: Any
+ ) -> ChatInvokeCompletion[str]: ...
@overload
- async def ainvoke(self, messages: list[BaseMessage], output_format: type[T]) -> ChatInvokeCompletion[T]: ...
+ async def ainvoke(self, messages: list[BaseMessage], output_format: type[T], **kwargs: Any) -> ChatInvokeCompletion[T]: ...
async def ainvoke(
- self, messages: list[BaseMessage], output_format: type[T] | None = None
+ self, messages: list[BaseMessage], output_format: type[T] | None = None, **kwargs: Any
) -> ChatInvokeCompletion[T] | ChatInvokeCompletion[str]:
"""
Invoke the AWS Bedrock model with the given messages.
diff --git a/browser_use/llm/base.py b/browser_use/llm/base.py
index 0e623d780..7a591c62f 100644
--- a/browser_use/llm/base.py
+++ b/browser_use/llm/base.py
@@ -32,13 +32,15 @@ class BaseChatModel(Protocol):
return self.model
@overload
- async def ainvoke(self, messages: list[BaseMessage], output_format: None = None) -> ChatInvokeCompletion[str]: ...
+ async def ainvoke(
+ self, messages: list[BaseMessage], output_format: None = None, **kwargs: Any
+ ) -> ChatInvokeCompletion[str]: ...
@overload
- async def ainvoke(self, messages: list[BaseMessage], output_format: type[T]) -> ChatInvokeCompletion[T]: ...
+ async def ainvoke(self, messages: list[BaseMessage], output_format: type[T], **kwargs: Any) -> ChatInvokeCompletion[T]: ...
async def ainvoke(
- self, messages: list[BaseMessage], output_format: type[T] | None = None
+ self, messages: list[BaseMessage], output_format: type[T] | None = None, **kwargs: Any
) -> ChatInvokeCompletion[T] | ChatInvokeCompletion[str]: ...
@classmethod
diff --git a/browser_use/llm/browser_use/chat.py b/browser_use/llm/browser_use/chat.py
index fcd5920b7..39224c00d 100644
--- a/browser_use/llm/browser_use/chat.py
+++ b/browser_use/llm/browser_use/chat.py
@@ -9,7 +9,7 @@ import asyncio
import logging
import os
import random
-from typing import TypeVar, overload
+from typing import Any, TypeVar, overload
import httpx
from pydantic import BaseModel
@@ -103,17 +103,21 @@ class ChatBrowserUse(BaseChatModel):
@overload
async def ainvoke(
- self, messages: list[BaseMessage], output_format: None = None, request_type: str = 'browser_agent'
+ self, messages: list[BaseMessage], output_format: None = None, request_type: str = 'browser_agent', **kwargs: Any
) -> ChatInvokeCompletion[str]: ...
@overload
async def ainvoke(
- self, messages: list[BaseMessage], output_format: type[T], request_type: str = 'browser_agent'
+ self, messages: list[BaseMessage], output_format: type[T], request_type: str = 'browser_agent', **kwargs: Any
) -> ChatInvokeCompletion[T]: ...
@observe(name='chat_browser_use_ainvoke')
async def ainvoke(
- self, messages: list[BaseMessage], output_format: type[T] | None = None, request_type: str = 'browser_agent'
+ self,
+ messages: list[BaseMessage],
+ output_format: type[T] | None = None,
+ request_type: str = 'browser_agent',
+ **kwargs: Any,
) -> ChatInvokeCompletion[T] | ChatInvokeCompletion[str]:
"""
Send request to browser-use cloud API.
@@ -122,6 +126,8 @@ class ChatBrowserUse(BaseChatModel):
messages: List of messages to send
output_format: Expected output format (Pydantic model)
request_type: Type of request - 'browser_agent' or 'judge'
+ **kwargs: Additional arguments, including:
+ - session_id: Session ID for sticky routing (same session → same container)
Returns:
ChatInvokeCompletion with structured response and usage info
@@ -131,8 +137,11 @@ class ChatBrowserUse(BaseChatModel):
anonymized_telemetry = CONFIG.ANONYMIZED_TELEMETRY
+ # Extract session_id from kwargs for sticky routing
+ session_id = kwargs.get('session_id')
+
# Prepare request payload
- payload = {
+ payload: dict[str, Any] = {
'model': self.model,
'messages': [self._serialize_message(msg) for msg in messages],
'fast': self.fast,
@@ -140,6 +149,10 @@ class ChatBrowserUse(BaseChatModel):
'anonymized_telemetry': anonymized_telemetry,
}
+ # Add session_id for sticky routing if provided
+ if session_id:
+ payload['session_id'] = session_id
+
# Add output format schema if provided
if output_format is not None:
payload['output_format'] = output_format.model_json_schema()
diff --git a/browser_use/llm/cerebras/chat.py b/browser_use/llm/cerebras/chat.py
index 6a3cb1f18..5df138f4a 100644
--- a/browser_use/llm/cerebras/chat.py
+++ b/browser_use/llm/cerebras/chat.py
@@ -77,6 +77,7 @@ class ChatCerebras(BaseChatModel):
self,
messages: list[BaseMessage],
output_format: None = None,
+ **kwargs: Any,
) -> ChatInvokeCompletion[str]: ...
@overload
@@ -84,12 +85,14 @@ class ChatCerebras(BaseChatModel):
self,
messages: list[BaseMessage],
output_format: type[T],
+ **kwargs: Any,
) -> ChatInvokeCompletion[T]: ...
async def ainvoke(
self,
messages: list[BaseMessage],
output_format: type[T] | None = None,
+ **kwargs: Any,
) -> ChatInvokeCompletion[T] | ChatInvokeCompletion[str]:
"""
Cerebras ainvoke supports:
diff --git a/browser_use/llm/deepseek/chat.py b/browser_use/llm/deepseek/chat.py
index b12a3744b..26f656c8b 100644
--- a/browser_use/llm/deepseek/chat.py
+++ b/browser_use/llm/deepseek/chat.py
@@ -66,6 +66,7 @@ class ChatDeepSeek(BaseChatModel):
output_format: None = None,
tools: list[dict[str, Any]] | None = None,
stop: list[str] | None = None,
+ **kwargs: Any,
) -> ChatInvokeCompletion[str]: ...
@overload
@@ -75,6 +76,7 @@ class ChatDeepSeek(BaseChatModel):
output_format: type[T],
tools: list[dict[str, Any]] | None = None,
stop: list[str] | None = None,
+ **kwargs: Any,
) -> ChatInvokeCompletion[T]: ...
async def ainvoke(
@@ -83,6 +85,7 @@ class ChatDeepSeek(BaseChatModel):
output_format: type[T] | None = None,
tools: list[dict[str, Any]] | None = None,
stop: list[str] | None = None,
+ **kwargs: Any,
) -> ChatInvokeCompletion[T] | ChatInvokeCompletion[str]:
"""
DeepSeek ainvoke supports:
diff --git a/browser_use/llm/google/chat.py b/browser_use/llm/google/chat.py
index 7334d3d37..c0b928502 100644
--- a/browser_use/llm/google/chat.py
+++ b/browser_use/llm/google/chat.py
@@ -184,13 +184,15 @@ class ChatGoogle(BaseChatModel):
return usage
@overload
- async def ainvoke(self, messages: list[BaseMessage], output_format: None = None) -> ChatInvokeCompletion[str]: ...
+ async def ainvoke(
+ self, messages: list[BaseMessage], output_format: None = None, **kwargs: Any
+ ) -> ChatInvokeCompletion[str]: ...
@overload
- async def ainvoke(self, messages: list[BaseMessage], output_format: type[T]) -> ChatInvokeCompletion[T]: ...
+ async def ainvoke(self, messages: list[BaseMessage], output_format: type[T], **kwargs: Any) -> ChatInvokeCompletion[T]: ...
async def ainvoke(
- self, messages: list[BaseMessage], output_format: type[T] | None = None
+ self, messages: list[BaseMessage], output_format: type[T] | None = None, **kwargs: Any
) -> ChatInvokeCompletion[T] | ChatInvokeCompletion[str]:
"""
Invoke the model with the given messages.
diff --git a/browser_use/llm/groq/chat.py b/browser_use/llm/groq/chat.py
index e973f538d..2ae10b9bd 100644
--- a/browser_use/llm/groq/chat.py
+++ b/browser_use/llm/groq/chat.py
@@ -1,6 +1,6 @@
import logging
from dataclasses import dataclass
-from typing import Literal, TypeVar, overload
+from typing import Any, Literal, TypeVar, overload
from groq import (
APIError,
@@ -100,13 +100,15 @@ class ChatGroq(BaseChatModel):
return usage
@overload
- async def ainvoke(self, messages: list[BaseMessage], output_format: None = None) -> ChatInvokeCompletion[str]: ...
+ async def ainvoke(
+ self, messages: list[BaseMessage], output_format: None = None, **kwargs: Any
+ ) -> ChatInvokeCompletion[str]: ...
@overload
- async def ainvoke(self, messages: list[BaseMessage], output_format: type[T]) -> ChatInvokeCompletion[T]: ...
+ async def ainvoke(self, messages: list[BaseMessage], output_format: type[T], **kwargs: Any) -> ChatInvokeCompletion[T]: ...
async def ainvoke(
- self, messages: list[BaseMessage], output_format: type[T] | None = None
+ self, messages: list[BaseMessage], output_format: type[T] | None = None, **kwargs: Any
) -> ChatInvokeCompletion[T] | ChatInvokeCompletion[str]:
groq_messages = GroqMessageSerializer.serialize_messages(messages)
diff --git a/browser_use/llm/mistral/chat.py b/browser_use/llm/mistral/chat.py
index ddd475647..f5ed47e31 100644
--- a/browser_use/llm/mistral/chat.py
+++ b/browser_use/llm/mistral/chat.py
@@ -159,13 +159,15 @@ class ChatMistral(BaseChatModel):
raise ModelProviderError(message=f'Failed to parse Mistral response: {e}', model=self.name) from e
@overload
- async def ainvoke(self, messages: list[BaseMessage], output_format: None = None) -> ChatInvokeCompletion[str]: ...
+ async def ainvoke(
+ self, messages: list[BaseMessage], output_format: None = None, **kwargs: Any
+ ) -> ChatInvokeCompletion[str]: ...
@overload
- async def ainvoke(self, messages: list[BaseMessage], output_format: type[T]) -> ChatInvokeCompletion[T]: ...
+ async def ainvoke(self, messages: list[BaseMessage], output_format: type[T], **kwargs: Any) -> ChatInvokeCompletion[T]: ...
async def ainvoke(
- self, messages: list[BaseMessage], output_format: type[T] | None = None
+ self, messages: list[BaseMessage], output_format: type[T] | None = None, **kwargs: Any
) -> ChatInvokeCompletion[T] | ChatInvokeCompletion[str]:
payload: dict[str, Any] = {
'model': self.model,
diff --git a/browser_use/llm/oci_raw/chat.py b/browser_use/llm/oci_raw/chat.py
index a0e60739a..7763471e9 100644
--- a/browser_use/llm/oci_raw/chat.py
+++ b/browser_use/llm/oci_raw/chat.py
@@ -8,7 +8,7 @@ Generative AI service using raw API calls without Langchain dependencies.
import asyncio
import json
from dataclasses import dataclass
-from typing import TypeVar, overload
+from typing import Any, TypeVar, overload
import oci
from oci.generative_ai_inference import GenerativeAiInferenceClient
@@ -322,13 +322,15 @@ class ChatOCIRaw(BaseChatModel):
return await loop.run_in_executor(None, _sync_request)
@overload
- async def ainvoke(self, messages: list[BaseMessage], output_format: None = None) -> ChatInvokeCompletion[str]: ...
+ async def ainvoke(
+ self, messages: list[BaseMessage], output_format: None = None, **kwargs: Any
+ ) -> ChatInvokeCompletion[str]: ...
@overload
- async def ainvoke(self, messages: list[BaseMessage], output_format: type[T]) -> ChatInvokeCompletion[T]: ...
+ async def ainvoke(self, messages: list[BaseMessage], output_format: type[T], **kwargs: Any) -> ChatInvokeCompletion[T]: ...
async def ainvoke(
- self, messages: list[BaseMessage], output_format: type[T] | None = None
+ self, messages: list[BaseMessage], output_format: type[T] | None = None, **kwargs: Any
) -> ChatInvokeCompletion[T] | ChatInvokeCompletion[str]:
"""
Invoke the OCI GenAI model with the given messages using raw API.
diff --git a/browser_use/llm/ollama/chat.py b/browser_use/llm/ollama/chat.py
index 99049b18a..0d592d448 100644
--- a/browser_use/llm/ollama/chat.py
+++ b/browser_use/llm/ollama/chat.py
@@ -58,13 +58,15 @@ class ChatOllama(BaseChatModel):
return self.model
@overload
- async def ainvoke(self, messages: list[BaseMessage], output_format: None = None) -> ChatInvokeCompletion[str]: ...
+ async def ainvoke(
+ self, messages: list[BaseMessage], output_format: None = None, **kwargs: Any
+ ) -> ChatInvokeCompletion[str]: ...
@overload
- async def ainvoke(self, messages: list[BaseMessage], output_format: type[T]) -> ChatInvokeCompletion[T]: ...
+ async def ainvoke(self, messages: list[BaseMessage], output_format: type[T], **kwargs: Any) -> ChatInvokeCompletion[T]: ...
async def ainvoke(
- self, messages: list[BaseMessage], output_format: type[T] | None = None
+ self, messages: list[BaseMessage], output_format: type[T] | None = None, **kwargs: Any
) -> ChatInvokeCompletion[T] | ChatInvokeCompletion[str]:
ollama_messages = OllamaMessageSerializer.serialize_messages(messages)
diff --git a/browser_use/llm/openai/chat.py b/browser_use/llm/openai/chat.py
index 367fb8eb1..ed363c9fc 100644
--- a/browser_use/llm/openai/chat.py
+++ b/browser_use/llm/openai/chat.py
@@ -146,13 +146,15 @@ class ChatOpenAI(BaseChatModel):
return usage
@overload
- async def ainvoke(self, messages: list[BaseMessage], output_format: None = None) -> ChatInvokeCompletion[str]: ...
+ async def ainvoke(
+ self, messages: list[BaseMessage], output_format: None = None, **kwargs: Any
+ ) -> ChatInvokeCompletion[str]: ...
@overload
- async def ainvoke(self, messages: list[BaseMessage], output_format: type[T]) -> ChatInvokeCompletion[T]: ...
+ async def ainvoke(self, messages: list[BaseMessage], output_format: type[T], **kwargs: Any) -> ChatInvokeCompletion[T]: ...
async def ainvoke(
- self, messages: list[BaseMessage], output_format: type[T] | None = None
+ self, messages: list[BaseMessage], output_format: type[T] | None = None, **kwargs: Any
) -> ChatInvokeCompletion[T] | ChatInvokeCompletion[str]:
"""
Invoke the model with the given messages.
diff --git a/browser_use/llm/openrouter/chat.py b/browser_use/llm/openrouter/chat.py
index cdbec29eb..f71fa4e06 100644
--- a/browser_use/llm/openrouter/chat.py
+++ b/browser_use/llm/openrouter/chat.py
@@ -114,13 +114,15 @@ class ChatOpenRouter(BaseChatModel):
)
@overload
- async def ainvoke(self, messages: list[BaseMessage], output_format: None = None) -> ChatInvokeCompletion[str]: ...
+ async def ainvoke(
+ self, messages: list[BaseMessage], output_format: None = None, **kwargs: Any
+ ) -> ChatInvokeCompletion[str]: ...
@overload
- async def ainvoke(self, messages: list[BaseMessage], output_format: type[T]) -> ChatInvokeCompletion[T]: ...
+ async def ainvoke(self, messages: list[BaseMessage], output_format: type[T], **kwargs: Any) -> ChatInvokeCompletion[T]: ...
async def ainvoke(
- self, messages: list[BaseMessage], output_format: type[T] | None = None
+ self, messages: list[BaseMessage], output_format: type[T] | None = None, **kwargs: Any
) -> ChatInvokeCompletion[T] | ChatInvokeCompletion[str]:
"""
Invoke the model with the given messages through OpenRouter.
diff --git a/browser_use/llm/vercel/chat.py b/browser_use/llm/vercel/chat.py
index 2d963d401..5ae0cd0dc 100644
--- a/browser_use/llm/vercel/chat.py
+++ b/browser_use/llm/vercel/chat.py
@@ -357,13 +357,15 @@ class ChatVercel(BaseChatModel):
return clean_schema(schema)
@overload
- async def ainvoke(self, messages: list[BaseMessage], output_format: None = None) -> ChatInvokeCompletion[str]: ...
+ async def ainvoke(
+ self, messages: list[BaseMessage], output_format: None = None, **kwargs: Any
+ ) -> ChatInvokeCompletion[str]: ...
@overload
- async def ainvoke(self, messages: list[BaseMessage], output_format: type[T]) -> ChatInvokeCompletion[T]: ...
+ async def ainvoke(self, messages: list[BaseMessage], output_format: type[T], **kwargs: Any) -> ChatInvokeCompletion[T]: ...
async def ainvoke(
- self, messages: list[BaseMessage], output_format: type[T] | None = None
+ self, messages: list[BaseMessage], output_format: type[T] | None = None, **kwargs: Any
) -> ChatInvokeCompletion[T] | ChatInvokeCompletion[str]:
"""
Invoke the model with the given messages through Vercel AI Gateway.