notte-agent 0.0.dev0__tar.gz

notte_agent-0.0.dev0/.gitignore

@@ -0,0 +1,179 @@
+# Byte-compiled / optimized / DLL files
+__pycache__/
+*.py[cod]
+*$py.class
+
+# C extensions
+*.so
+
+# Distribution / packaging
+.Python
+build/
+develop-eggs/
+dist/
+downloads/
+eggs/
+.eggs/
+lib/
+lib64/
+parts/
+sdist/
+var/
+wheels/
+share/python-wheels/
+*.egg-info/
+.installed.cfg
+*.egg
+MANIFEST
+
+# PyInstaller
+#  Usually these files are written by a python script from a template
+#  before PyInstaller builds the exe, so as to inject date/other infos into it.
+*.manifest
+*.spec
+
+# Installer logs
+pip-log.txt
+pip-delete-this-directory.txt
+
+# Unit test / coverage reports
+htmlcov/
+.tox/
+.nox/
+.coverage
+.coverage.*
+.cache
+nosetests.xml
+coverage.xml
+*.cover
+*.py,cover
+.hypothesis/
+.pytest_cache/
+cover/
+
+# Translations
+*.mo
+*.pot
+
+# Django stuff:
+*.log
+local_settings.py
+db.sqlite3
+db.sqlite3-journal
+
+# Flask stuff:
+instance/
+.webassets-cache
+
+# Scrapy stuff:
+.scrapy
+
+# Sphinx documentation
+docs/_build/
+
+# PyBuilder
+.pybuilder/
+target/
+
+# Jupyter Notebook
+.ipynb_checkpoints
+
+# IPython
+profile_default/
+ipython_config.py
+
+# pyenv
+#   For a library or package, you might want to ignore these files since the code is
+#   intended to run in multiple environments; otherwise, check them in:
+# .python-version
+
+# pipenv
+#   According to pypa/pipenv#598, it is recommended to include Pipfile.lock in version control.
+#   However, in case of collaboration, if having platform-specific dependencies or dependencies
+#   having no cross-platform support, pipenv may install dependencies that don't work, or not
+#   install all needed dependencies.
+#Pipfile.lock
+
+# UV
+#   Similar to Pipfile.lock, it is generally recommended to include uv.lock in version control.
+#   This is especially recommended for binary packages to ensure reproducibility, and is more
+#   commonly ignored for libraries.
+#uv.lock
+
+# poetry
+#   Similar to Pipfile.lock, it is generally recommended to include poetry.lock in version control.
+#   This is especially recommended for binary packages to ensure reproducibility, and is more
+#   commonly ignored for libraries.
+#   https://python-poetry.org/docs/basic-usage/#commit-your-poetrylock-file-to-version-control
+#poetry.lock
+
+# pdm
+#   Similar to Pipfile.lock, it is generally recommended to include pdm.lock in version control.
+#pdm.lock
+#   pdm stores project-wide configurations in .pdm.toml, but it is recommended to not include it
+#   in version control.
+#   https://pdm.fming.dev/latest/usage/project/#working-with-version-control
+.pdm.toml
+.pdm-python
+.pdm-build/
+
+# PEP 582; used by e.g. github.com/David-OConnor/pyflow and github.com/pdm-project/pdm
+__pypackages__/
+
+# Celery stuff
+celerybeat-schedule
+celerybeat.pid
+
+# SageMath parsed files
+*.sage.py
+
+# Environments
+.env
+.venv
+env/
+venv/
+ENV/
+env.bak/
+venv.bak/
+
+# Spyder project settings
+.spyderproject
+.spyproject
+
+# Rope project settings
+.ropeproject
+
+# mkdocs documentation
+/site
+
+# mypy
+.mypy_cache/
+.dmypy.json
+dmypy.json
+
+# Pyre type checker
+.pyre/
+
+# pytype static type analyzer
+.pytype/
+
+# Cython debug symbols
+cython_debug/
+
+# PyCharm
+#  JetBrains specific template is maintained in a separate JetBrains.gitignore that can
+#  be found at https://github.com/github/gitignore/blob/main/Global/JetBrains.gitignore
+#  and can be added to the global gitignore or merged into this file.  For a more nuclear
+#  option (not recommended) you can uncomment the following to ignore the entire idea folder.
+#.idea/
+
+ignore.*
+llm_usage.jsonl
+llm_parsing_error.jsonl
+traces/
+
+**/__pycache__/**
+.DS_Store
+**/.DS_Store
+old
+notebook

notte_agent-0.0.dev0/PKG-INFO

@@ -0,0 +1,8 @@
+Metadata-Version: 2.4
+Name: notte-agent
+Version: 0.0.dev0
+Summary: Notte Web AI Agents
+Author-email: Notte Team  <hello@notte.cc>
+Requires-Python: >=3.11
+Requires-Dist: notte-browser==0.0.dev
+Requires-Dist: notte-core==0.0.dev

notte_agent-0.0.dev0/pyproject.toml

@@ -0,0 +1,22 @@
+[project]
+name = "notte-agent"
+version = "0.0.dev"
+description = "Notte Web AI Agents"
+readme = "README.md"
+authors = [
+    { name = "Notte Team ", email = "hello@notte.cc" }
+]
+packages = [
+  { include = "notte_agent", from = "src" },
+]
+
+requires-python = ">=3.11"
+dependencies = [
+    "notte_core==0.0.dev",
+    "notte_browser==0.0.dev",
+]
+
+
+[build-system]
+requires = ["hatchling"]
+build-backend = "hatchling.build"

notte_agent-0.0.dev0/src/notte_agent/README.md

@@ -0,0 +1,58 @@
+# How to build an LLM agent with *Notte*
+
+This guide explains how to build a custom LLM agent using *Notte*. The example in `agent.py` demonstrates a basic implementation that you can customize for your specific needs.
+
+## Overview
+
+*Notte* provides a flexible environment for web automation that can be controlled through an API. To build an agent with *Notte*, you need:
+
+1. An agent implementation that coordinates between your LLM and the *Notte* environment
+2. A parser that formats *Notte*'s outputs into prompts suitable for your LLM
+3. A way to interpret the LLM's responses back into *Notte* commands
+
+## Key Components
+
+### Agent
+
+The `Agent` class in `agent.py` shows how to:
+- Initialize a connection to your LLM service
+- Manage the conversation flow between the LLM and *Notte*
+- Track the state of task completion
+
+### Parser
+
+The parser is crucial for translating between *Notte* and your LLM. You'll need to:
+
+1. Create a custom parser (by extending `BaseNotteParser` or implementing the `Parser` interface)
+2. Define how to format:
+   - Observations from web pages
+   - Available actions
+   - Data extraction results
+   - Task completion status
+
+The provided `BaseNotteParser` is a simple example that you should modify based on your needs. Consider:
+- The prompt format your LLM works best with
+- How to structure web observations for your specific tasks
+- What action format makes sense for your use case
+- How to handle task completion and data extraction
+
+## Example Implementation
+
+See `agent.py` for a basic implementation. Key points to customize:
+- The parser implementation
+- The prompt engineering in the conversation flow
+- How task completion is determined
+- Error handling and retry logic
+
+## Best Practices
+
+1. **Custom Parser**: Don't just use the `BaseNotteParser` as-is. Create your own parser that:
+   - Formats observations in a way that makes sense for your LLM
+   - Structures action possibilities clearly
+   - Handles task-specific data extraction
+
+2. **Prompt Engineering**: Carefully design your system prompt and conversation flow
+
+3. **Error Handling**: Add robust error handling for both LLM and *Notte* interactions
+
+4. **Testing**: Test your parser and agent with different scenarios

notte_agent-0.0.dev0/src/notte_agent/__init__.py

@@ -0,0 +1,7 @@
+from notte_core import check_notte_version
+
+from notte_agent.main import Agent
+
+__version__ = check_notte_version("notte_agent")
+
+__all__ = ["Agent"]

notte_agent-0.0.dev0/src/notte_agent/common/base.py

@@ -0,0 +1,14 @@
+from abc import ABC, abstractmethod
+
+from notte_browser.session import NotteSession
+
+from notte_agent.common.types import AgentResponse
+
+
+class BaseAgent(ABC):
+    def __init__(self, session: NotteSession):
+        self.session: NotteSession = session
+
+    @abstractmethod
+    async def run(self, task: str, url: str | None = None) -> AgentResponse:
+        pass

notte_agent-0.0.dev0/src/notte_agent/common/captcha_detector.py

@@ -0,0 +1,87 @@
+from typing import final
+
+import chevron
+from notte_browser.session import TrajectoryStep
+from notte_core.llms.engine import LLMEngine
+from pydantic import BaseModel
+
+from notte_agent.common.conversation import Conversation
+from notte_agent.common.perception import BasePerception
+
+system_rules = """
+You are a captcha detector for web pages.
+Analyze the provided screenshot and determine if there is a captcha present on the page.
+A captcha can be in various forms such as:
+- Image-based challenges
+- Text-based challenges
+- Checkbox-based verification
+- Audio-based challenges
+- Math problems
+- Slider puzzles
+- Blocked by network security
+- etc.
+
+Return a JSON object with 3 keys: `has_captcha`, `captcha_type`, and `description`:
+- `has_captcha` is a boolean indicating if a captcha is present
+- `captcha_type` is a string describing the type of captcha (or "none" if no captcha)
+- `description` is a string providing details about the captcha's appearance and location
+
+Example:
+```json
+{{&example}}
+```
+
+Your turn:
+"""
+
+
+class CaptchaDetection(BaseModel):
+    has_captcha: bool
+    captcha_type: str
+    description: str
+
+
+@final
+class CaptchaDetector:
+    def __init__(
+        self,
+        llm: LLMEngine,
+        perception: BasePerception,
+        use_vision: bool = True,
+        include_attributes: bool = True,
+    ):
+        self.use_vision = use_vision
+        self.include_attributes = include_attributes
+        self.llm: LLMEngine = llm
+        self.conv: Conversation = Conversation()
+        self.perception: BasePerception = perception
+
+    @staticmethod
+    def example() -> CaptchaDetection:
+        return CaptchaDetection(
+            has_captcha=True,
+            captcha_type="image-based",
+            description="A grid of 9 images is shown with the instruction 'Select all images containing traffic lights'",
+        )
+
+    def detection_message(self, step: TrajectoryStep) -> str:
+        return f"""
+Current page screenshot:
+{self.perception.perceive(step.obs)}
+
+Current page state:
+{step.action.model_dump_json(exclude_unset=True)}
+"""
+
+    def detect(
+        self,
+        step: TrajectoryStep,
+    ) -> CaptchaDetection:
+        """Detect if there is a captcha present in the current page screenshot"""
+        self.conv.reset()
+        system_prompt = chevron.render(system_rules, {"example": self.example().model_dump_json()})
+        self.conv.add_system_message(content=system_prompt)
+        self.conv.add_user_message(content=self.detection_message(step))
+
+        answer: CaptchaDetection = self.llm.structured_completion(self.conv.messages(), CaptchaDetection)
+        return answer

notte_agent-0.0.dev0/src/notte_agent/common/config.py

@@ -0,0 +1,219 @@
+from abc import ABC, abstractmethod
+from argparse import ArgumentParser, Namespace
+from collections.abc import Callable
+from enum import StrEnum
+from typing import Any, ClassVar, Self, get_origin, get_type_hints
+
+from notte_browser.session import NotteSessionConfig
+from notte_core.common.config import FrozenConfig
+from notte_core.llms.engine import LlmModel
+from notte_sdk.types import DEFAULT_MAX_NB_STEPS
+from pydantic import Field, model_validator
+
+
+class RaiseCondition(StrEnum):
+    """How to raise an error when the agent fails to complete a step.
+
+    Either immediately upon failure, after retry, or never.
+    """
+
+    IMMEDIATELY = "immediately"
+    RETRY = "retry"
+    NEVER = "never"
+
+
+class DefaultAgentArgs(StrEnum):
+    SESSION_DISABLE_WEB_SECURITY = "disable_web_security"
+    SESSION_HEADLESS = "headless"
+    SESSION_PERCEPTION_MODEL = "perception_model"
+    SESSION_MAX_STEPS = "max_steps"
+
+    def with_prefix(self: Self, prefix: str = "session") -> str:
+        return f"{prefix}.{self.value}"
+
+
+class AgentConfig(FrozenConfig, ABC):
+    # make session private to avoid exposing the NotteSessionConfig class
+    session: NotteSessionConfig = Field(init=False)
+    reasoning_model: str = Field(
+        default=LlmModel.default(), description="The model to use for reasoning (i.e taking actions)."
+    )
+    include_screenshot: bool = Field(default=False, description="Whether to include a screenshot in the response.")
+    max_history_tokens: int | None = Field(
+        default=None,
+        description="The maximum number of tokens in the history. When the history exceeds this limit, the oldest messages are discarded.",
+    )
+    max_error_length: int = Field(
+        default=500, description="The maximum length of an error message to be forwarded to the reasoning model."
+    )
+    raise_condition: RaiseCondition = Field(
+        default=RaiseCondition.RETRY, description="How to raise an error when the agent fails to complete a step."
+    )
+    max_consecutive_failures: int = Field(
+        default=3, description="The maximum number of consecutive failures before the agent gives up."
+    )
+    force_session: bool | None = Field(
+        default=None,
+        description="Whether to allow the user to set the session or not.",
+    )
+    human_in_the_loop: bool = Field(default=False, description="Whether to enable human-in-the-loop mode.")
+
+    @classmethod
+    @abstractmethod
+    def default_session(cls) -> NotteSessionConfig:
+        raise NotImplementedError("Subclasses must implement this method")
+
+    @model_validator(mode="before")
+    @classmethod
+    def set_session(cls, values: dict[str, Any]) -> dict[str, Any]:
+        if "session" in values:
+            if "force_session" in values and values["force_session"]:
+                del values["force_session"]
+                return values
+            raise ValueError("Session should not be set by the user. Set `default_session` instead.")
+        values["session"] = cls.default_session()  # Set the session field using the subclass's method
+        return values
+
+    def groq(self: Self, deep: bool = True) -> Self:
+        return self.model(LlmModel.groq, deep=deep)
+
+    def openai(self: Self, deep: bool = True) -> Self:
+        return self.model(LlmModel.openai, deep=deep)
+
+    def gemini(self: Self, deep: bool = True) -> Self:
+        return self.model(LlmModel.gemini, deep=deep)
+
+    def cerebras(self: Self, deep: bool = True) -> Self:
+        return self.model(LlmModel.cerebras, deep=deep)
+
+    def model(self: Self, model: LlmModel, deep: bool = True) -> Self:
+        config = self._copy_and_validate(reasoning_model=model, max_history_tokens=LlmModel.context_length(model))
+        if deep:
+            config = config.map_session(lambda session: session.model(model))
+        return config
+
+    def use_vision(self: Self, value: bool = True) -> Self:
+        return self._copy_and_validate(include_screenshot=value)
+
+    def set_human_in_the_loop(self: Self, value: bool = True) -> Self:
+        return self._copy_and_validate(human_in_the_loop=value)
+
+    def dev_mode(self: Self) -> Self:
+        return self._copy_and_validate(
+            raise_condition=RaiseCondition.IMMEDIATELY,
+            max_error_length=1000,
+            session=self.session.dev_mode(),
+            force_session=True,
+        )
+
+    def set_raise_condition(self: Self, value: RaiseCondition) -> Self:
+        return self._copy_and_validate(raise_condition=value)
+
+    def map_session(self: Self, ft: Callable[[NotteSessionConfig], NotteSessionConfig]) -> Self:
+        return self._copy_and_validate(session=ft(self.session), force_session=True)
+
+    @staticmethod
+    def _get_arg_type(python_type: Any) -> Any:
+        """Maps Python types to argparse types."""
+        type_map = {
+            str: str,
+            int: int,
+            float: float,
+            bool: bool,
+        }
+        return type_map.get(python_type, str)
+
+    @staticmethod
+    def create_base_parser() -> ArgumentParser:
+        """Creates a base ArgumentParser with all the fields from the config."""
+        parser = ArgumentParser()
+        _ = parser.add_argument(
+            f"--{DefaultAgentArgs.SESSION_HEADLESS.with_prefix()}",
+            action="store_true",
+            help="Whether to run the browser in headless mode.",
+        )
+        _ = parser.add_argument(
+            f"--{DefaultAgentArgs.SESSION_DISABLE_WEB_SECURITY.with_prefix()}",
+            action="store_true",
+            help="Whether disable web security.",
+        )
+        _ = parser.add_argument(
+            f"--{DefaultAgentArgs.SESSION_PERCEPTION_MODEL.with_prefix()}",
+            type=str,
+            default=LlmModel.default(),
+            help="The model to use for perception.",
+        )
+        _ = parser.add_argument(
+            f"--{DefaultAgentArgs.SESSION_MAX_STEPS.with_prefix()}",
+            type=int,
+            default=DEFAULT_MAX_NB_STEPS,
+            help="The maximum number of steps the agent can take.",
+        )
+        return parser
+
+    @classmethod
+    def create_parser(cls) -> ArgumentParser:
+        """Creates an ArgumentParser with all the fields from the config."""
+        parser = cls.create_base_parser()
+        hints = get_type_hints(cls)
+
+        for field_name, field_info in cls.model_fields.items():
+            if field_name == "session":
+                continue
+            field_type = hints.get(field_name)
+            if get_origin(field_type) is ClassVar:
+                continue
+
+            default = field_info.default
+            help_text = field_info.description or "no description available"
+            arg_type = cls._get_arg_type(field_type)
+
+            _ = parser.add_argument(
+                f"--{field_name.replace('_', '-')}",
+                type=arg_type,
+                default=default,
+                help=f"{help_text} (default: {default})",
+            )
+
+        return parser
+
+    @classmethod
+    def from_args(cls: type[Self], args: Namespace) -> Self:
+        """Creates an AgentConfig from a Namespace of arguments.
+
+        The return type will match the class that called this method.
+        """
+        disallowed_args = ["task", "session.window.headless"]
+
+        session_args = {
+            k.replace("session.", "").replace("-", "_"): v
+            for k, v in vars(args).items()
+            if k.startswith("session.") and k not in disallowed_args
+        }
+        agent_args = {
+            k.replace("-", "_"): v
+            for k, v in vars(args).items()
+            if not k.startswith("session.") and k not in disallowed_args
+        }
+
+        def update_session(session: NotteSessionConfig) -> NotteSessionConfig:
+            operations: list[Callable[[NotteSessionConfig], NotteSessionConfig]] = []
+            if DefaultAgentArgs.SESSION_HEADLESS in session_args:
+                headless = session_args[DefaultAgentArgs.SESSION_HEADLESS]
+                operations.append(lambda session: session.headless(headless))
+                del session_args[DefaultAgentArgs.SESSION_HEADLESS]
+            if DefaultAgentArgs.SESSION_DISABLE_WEB_SECURITY in session_args:
+                disable_web_security = session_args[DefaultAgentArgs.SESSION_DISABLE_WEB_SECURITY]
+                operations.append(
+                    lambda session: session.disable_web_security()
+                    if disable_web_security
+                    else session.enable_web_security()
+                )
+                del session_args[DefaultAgentArgs.SESSION_DISABLE_WEB_SECURITY]
+
+            session = session._copy_and_validate(**session_args)
+            for operation in operations:
+                session = operation(session)
+            return session
+
+        return cls(**agent_args).map_session(update_session)