rasa-pro 3.14.0.dev5__py3-none-any.whl → 3.14.0.dev7__py3-none-any.whl

rasa/builder/download.py

@@ -1,140 +0,0 @@
-"""Download utilities for bot projects."""
-
-import io
-import os
-import sys
-import tarfile
-from textwrap import dedent
-from typing import Dict, Optional
-
-
-def _get_env_content() -> str:
-    """Generate .env file content."""
-    return f"RASA_PRO_LICENSE={os.getenv('RASA_PRO_LICENSE')}\n"
-
-
-def _get_python_version_content() -> str:
-    """Generate .python-version file content with current Python version."""
-    return f"{sys.version_info.major}.{sys.version_info.minor}\n"
-
-
-def _get_pyproject_toml_content(project_id: str) -> str:
-    """Generate pyproject.toml file content."""
-    return dedent(
-        f"""
-        [project]
-        name = "{project_id}"
-        version = "0.1.0"
-        description = "Add your description for your Rasa bot here"
-        readme = "README.md"
-        dependencies = []
-        requires-python = ">={sys.version_info.major}.{sys.version_info.minor}"
-        """
-    )
-
-
-def _get_readme_content() -> str:
-    """Generate README.md file content with basic instructions."""
-    return dedent(
-        """
-        # Rasa Assistant
-
-        This is your Rasa assistant project. Below are the basic commands to
-        get started.
-
-        ## Prerequisites
-        Make sure you have [uv](https://docs.astral.sh/uv/) installed.
-
-        ## Training the Assistant
-
-        To train your assistant with the current configuration and training data:
-
-        ```bash
-        uv run rasa train
-        ```
-
-        This will create a model in the `models/` directory.
-
-        ## Testing the Assistant
-
-        To test your assistant interactively in the command line:
-
-        ```bash
-        uv run rasa inspect
-        ```
-
-        ## Running the Assistant
-
-        To start the Rasa server:
-
-        ```bash
-        uv run rasa run
-        ```
-
-        The server will be available at `http://localhost:5005`.
-
-        ## Project Structure
-
-        - `config.yml` - Configuration for your NLU pipeline and policies
-        - `domain.yml` - Defines intents, entities, slots, responses, and actions
-        - `data/` - Flows of your bot
-        - `actions/` - Custom action code (if any)
-
-        ## Next Steps
-
-        1. Customize your domain and flows
-        2. Train your model with `rasa train`
-        3. Test your assistant with `rasa inspect`
-        4. Deploy your assistant with `rasa run`
-
-        For more information, visit the [Rasa documentation](https://rasa.com/docs/).
-        """
-    )
-
-
-def _add_file_to_tar(tar: tarfile.TarFile, filename: str, content: str) -> None:
-    """Add a file with the given content to the tar archive.
-
-    Args:
-        tar: The tar file object to add to
-        filename: Name of the file in the archive
-        content: Content of the file as a string
-    """
-    file_data = content.encode("utf-8")
-    tarinfo = tarfile.TarInfo(name=filename)
-    tarinfo.size = len(file_data)
-    tar.addfile(tarinfo, io.BytesIO(file_data))
-
-
-def create_bot_project_archive(
-    bot_files: Dict[str, Optional[str]], project_id: str
-) -> bytes:
-    """Create a tar.gz archive containing bot files and additional project files.
-
-    Args:
-        bot_files: Dictionary mapping file names to their content
-        project_id: Name of the project for the archive filename and pyproject.toml
-
-    Returns:
-        bytes: The tar.gz archive data
-    """
-    tar_buffer = io.BytesIO()
-
-    with tarfile.open(fileobj=tar_buffer, mode="w:gz") as tar:
-        # Add bot files to archive
-        for filename, content in bot_files.items():
-            if content is not None:
-                _add_file_to_tar(tar, filename, content)
-
-        # Add additional project files
-        _add_file_to_tar(tar, ".env", _get_env_content())
-        _add_file_to_tar(tar, ".python-version", _get_python_version_content())
-        _add_file_to_tar(
-            tar,
-            "pyproject.toml",
-            _get_pyproject_toml_content(project_id),
-        )
-        _add_file_to_tar(tar, "README.md", _get_readme_content())
-
-    tar_buffer.seek(0)
-    return tar_buffer.getvalue()

rasa/builder/exceptions.py

@@ -1,55 +0,0 @@
-"""Custom exceptions for the prompt-to-bot service."""
-
-from typing import Any, Optional
-
-
-class PromptToBotError(Exception):
-    """Base exception for prompt-to-bot service."""
-
-    pass
-
-
-class ValidationError(PromptToBotError):
-    """Raised when Rasa project validation fails."""
-
-    def __init__(self, message: str, validation_logs: Optional[Any] = None):
-        super().__init__(message)
-        self.validation_logs = validation_logs
-
-
-class TrainingError(PromptToBotError):
-    """Raised when model training fails."""
-
-    pass
-
-
-class LLMGenerationError(PromptToBotError):
-    """Raised when LLM generation fails."""
-
-    pass
-
-
-class DocumentRetrievalError(PromptToBotError):
-    """Raised when document retrieval fails."""
-
-    pass
-
-
-class SchemaValidationError(PromptToBotError):
-    """Raised when schema validation fails."""
-
-    pass
-
-
-class AgentLoadError(PromptToBotError):
-    """Raised when agent loading fails."""
-
-    pass
-
-
-class ProjectGenerationError(PromptToBotError):
-    """Raised when project generation fails after retries."""
-
-    def __init__(self, message: str, attempts: int):
-        super().__init__(f"{message} (failed after {attempts} attempts)")
-        self.attempts = attempts

rasa/builder/guardrails/__init__.py

@@ -1 +0,0 @@
-

rasa/builder/guardrails/constants.py

@@ -1,3 +0,0 @@
-LAKERA_API_KEY_ENV_VAR = "LAKERA_API_KEY"
-LAKERA_GUARD_ENDPOINT = "guard"
-LAKERA_GUARD_RESULTS_ENDPOINT = "guard/results"

rasa/builder/guardrails/exceptions.py

@@ -1,4 +0,0 @@
-class GuardrailsError(Exception):
-    """Raised when a guardrails error occurs."""
-
-    pass

rasa/builder/guardrails/lakera.py

@@ -1,206 +0,0 @@
-"""Lakera AI guardrails provider implementation."""
-
-import asyncio
-import os
-import time
-from contextlib import asynccontextmanager
-from typing import Any, AsyncGenerator, Dict, Optional
-
-import aiohttp
-import structlog
-
-from rasa.builder import config
-from rasa.builder.guardrails.constants import (
-    LAKERA_API_KEY_ENV_VAR,
-    LAKERA_GUARD_ENDPOINT,
-    LAKERA_GUARD_RESULTS_ENDPOINT,
-)
-from rasa.builder.guardrails.exceptions import GuardrailsError
-from rasa.builder.guardrails.models import (
-    GuardrailResponse,
-    LakeraGuardrailRequest,
-    LakeraGuardrailResponse,
-)
-
-structlogger = structlog.get_logger()
-
-
-class LakeraAIGuardrails:
-    """Guardrails provider using Lakera AI."""
-
-    def __init__(
-        self,
-        api_key: Optional[str] = None,
-        base_url: Optional[str] = None,
-    ):
-        """Initialize Lakera guardrails provider.
-
-        Args:
-            severity_threshold: The minimum severity level for the guardrail to flag
-                the user request as unsafe.
-            api_key: Lakera AI API key.
-            base_url: Optional base URL for the API. If not provided, the default
-                Lakera API URL (https://api.lakera.ai/v2) will be used.
-        """
-        self.base_url: str = base_url or config.LAKERA_BASE_URL
-        self._api_key: Optional[str] = api_key or os.getenv(LAKERA_API_KEY_ENV_VAR)
-        self._session: Optional[aiohttp.ClientSession] = None
-
-    @property
-    def guard_endpoint(self) -> str:
-        """Get the guard endpoint for the Lakera API."""
-        return f"{self.base_url}/{LAKERA_GUARD_ENDPOINT}"
-
-    @property
-    def guard_results_endpoint(self) -> str:
-        """Get the guard results endpoint for the Lakera API."""
-        return f"{self.base_url}/{LAKERA_GUARD_RESULTS_ENDPOINT}"
-
-    @asynccontextmanager
-    async def _get_session(self) -> AsyncGenerator[aiohttp.ClientSession, None]:
-        """Create a fresh ClientSession, yield it, and always close it."""
-        session = aiohttp.ClientSession(headers=self._get_headers())
-        structlogger.debug("lakera_guardrails._get_session", base_url=self.base_url)
-
-        try:
-            yield session
-        except Exception as e:
-            structlogger.error("lakera_guardrails.session_error", error=str(e))
-            raise
-        finally:
-            try:
-                await session.close()
-            except Exception as exc:
-                structlogger.warning(
-                    "lakera_guardrails.session_close_error",
-                    event_info="Failed to close aiohttp client session cleanly.",
-                    error=str(exc),
-                )
-
-    def _get_headers(self) -> Dict[str, str]:
-        """Get the headers for the Lakera API request.
-
-        Returns:
-            A dictionary containing the Authorization header with the API key.
-        """
-        using_proxy = bool(config.HELLO_LLM_PROXY_BASE_URL)
-
-        if using_proxy:
-            if not config.RASA_PRO_LICENSE:
-                raise GuardrailsError(
-                    "HELLO_LLM_PROXY_BASE_URL is set but RASA_PRO_LICENSE is missing. "
-                    "Proxy requires a Rasa Pro license token for authentication."
-                )
-            return {"Authorization": f"Bearer {config.RASA_PRO_LICENSE}"}
-
-        if not self._api_key:
-            raise GuardrailsError(
-                "LAKERA_API_KEY is missing. Provide it via env LAKERA_API_KEY or "
-                "pass api_key= to LakeraAIGuardrails."
-            )
-        return {"Authorization": f"Bearer {self._api_key}"}
-
-    async def send_request(self, request: LakeraGuardrailRequest) -> GuardrailResponse:
-        """Send a request to the Lakera API.
-
-        Args:
-            request: The guardrail request to send to the Lakera API.
-
-        Returns:
-            GuardrailResponse with the results of the check.
-
-        Raises:
-            GuardrailsError: If the request times out or returns a non-200 status code.
-            Exception: If the request fails for any other reason.
-        """
-        start_time = time.time()
-        try:
-            async with self._get_session() as session:
-                raw_response = await self._send_http_request(session, request)
-                response = LakeraGuardrailResponse.from_raw_response(
-                    raw_response,
-                    hello_rasa_user_id=request.hello_rasa_user_id,
-                    hello_rasa_project_id=request.hello_rasa_project_id,
-                )
-                processing_time_ms = (time.time() - start_time) * 1000
-                response.processing_time_ms = processing_time_ms
-                return response
-
-        # Propagate the GuardrailsError if it occurs.
-        except GuardrailsError as e:
-            raise e
-
-        except asyncio.TimeoutError:
-            message = "Lakera API request timed out."
-            structlogger.error(
-                "lakera_guardrails.send_request.timeout_error",
-                event_info=message,
-                processing_time_ms=(time.time() - start_time) * 1000,
-            )
-            raise GuardrailsError(message)
-
-        # Propagate the unexpected exceptions.
-        except Exception as e:
-            message = "Lakera API request failed."
-            structlogger.error(
-                "lakera_guardrails.send_request.unexpected_error",
-                event_info="Lakera API request failed.",
-                request=request,
-                error=e,
-                processing_time_ms=(time.time() - start_time) * 1000,
-            )
-            raise e
-
-    async def _send_http_request(
-        self, session: aiohttp.ClientSession, request: LakeraGuardrailRequest
-    ) -> Dict[str, Any]:
-        """Make an HTTP request to the Lakera API.
-
-        Args:
-            session: The aiohttp session to use for the request.
-            request: The guardrail request to send.
-
-        Returns:
-            The raw JSON response from the API.
-
-        Raises:
-            GuardrailsError: If the request fails or response parsing fails.
-        """
-        # Log the request details for debugging
-        json_payload = request.to_json_payload()
-        structlogger.debug(
-            "lakera_guardrails.send_request.request",
-            url=self.guard_endpoint,
-            method="POST",
-            request_body=json_payload,
-        )
-
-        async with session.post(
-            self.guard_endpoint,
-            json=json_payload,
-        ) as client_response:
-            # Check if the response is successful. If not, raise an error.
-            if client_response.status >= 400:
-                error_text = await client_response.text()
-                message = (
-                    f"Lakera API request failed with status "
-                    f"`{client_response.status}`. Error: "
-                    f"`{error_text}`."
-                )
-                structlogger.error(
-                    "lakera_guardrails.send_request.http_error",
-                    event_info=message,
-                    url=self.guard_endpoint,
-                    status=client_response.status,
-                    error=error_text,
-                    request_body=json_payload,
-                )
-                raise GuardrailsError(message)
-
-            # Parse the response as a dictionary.
-            raw_response = await client_response.json()
-            structlogger.debug(
-                "lakera_guardrails.send_request.response",
-                response_body=raw_response,
-            )
-            return raw_response

rasa/builder/guardrails/models.py

@@ -1,199 +0,0 @@
-"""Models for guardrails system."""
-
-import copy
-from abc import ABC, abstractmethod
-from enum import Enum
-from typing import Any, Dict, List, Optional
-
-from pydantic import BaseModel, ConfigDict, Field
-
-
-class GuardrailType(Enum):
-    """Types of guardrails that can be applied with Lakera AI."""
-
-    PROMPT_ATTACK = "prompt_attack"
-    CONTENT_VIOLATION = "content_violation"
-    DATA_LEAKAGE = "data_leakage"
-    MALICIOUS_LINKS = "malicious_content"
-    CUSTOM = "custom"
-    OTHER = "other"
-
-
-class GuardrailRequest(BaseModel, ABC):
-    """Request for guardrails check."""
-
-    hello_rasa_user_id: Optional[str] = Field(
-        default=None,
-        description="Required. User identifier for the Hello Rasa project. ",
-    )
-    hello_rasa_project_id: Optional[str] = Field(
-        default=None,
-        description="Required. Project identifier for the Hello Rasa project. ",
-    )
-    metadata: Optional[Dict[str, Any]] = Field(
-        default=None, description="Additional metadata for the guardrails endpoint."
-    )
-
-    @abstractmethod
-    def to_json_payload(self) -> Dict[str, Any]:
-        """Convert the request to a JSON payload."""
-        ...
-
-
-class LakeraGuardrailRequest(GuardrailRequest):
-    """Request for Lakera guardrails check."""
-
-    lakera_project_id: str = Field(
-        description="Required. Project identifier for the Lakera AI project."
-    )
-    payload: bool = Field(
-        default=True,
-        description=(
-            "From Lakera AI: When true the response will return a payload object "
-            "containing any PII, profanity or custom detector regex matches detected, "
-            "along with their location within the contents."
-        ),
-    )
-    breakdown: bool = Field(
-        default=True,
-        description=(
-            "From Lakera AI: When true the response will return a breakdown list of "
-            "the detectors that were run, as defined in the policy, and whether each "
-            "of them detected something or not."
-        ),
-    )
-
-    messages: List[Dict[str, Any]] = Field(
-        description=(
-            "Required. From Lakera AI: List of messages comprising the interaction "
-            "history with the LLM in OpenAI API Chat Completions format. Can be "
-            "multiple messages of any role: user, assistant, system, tool, or "
-            "developer."
-        ),
-    )
-
-    def to_json_payload(self) -> Dict[str, Any]:
-        """Convert the request to a JSON payload to be sent to the Lakera endpoint."""
-        metadata = self.metadata or {}
-        metadata["hello_rasa_project_id"] = self.hello_rasa_project_id
-        metadata["hello_rasa_user_id"] = self.hello_rasa_user_id
-
-        json_payload: Dict[str, Any] = {
-            "messages": self.messages,
-            "project_id": self.lakera_project_id,
-            "metadata": metadata,
-        }
-
-        if self.payload:
-            json_payload["payload"] = self.payload
-        if self.breakdown:
-            json_payload["breakdown"] = self.breakdown
-
-        return json_payload
-
-
-class GuardrailDetection(BaseModel):
-    """Represents a single guardrail detection."""
-
-    type: GuardrailType = Field(description="Type of guardrail detection.")
-    original_type: str = Field(description="Original detection from the provider.")
-    metadata: Optional[Dict[str, Any]] = Field(
-        default=None, description="Additional metadata about the detection itself."
-    )
-
-
-class GuardrailResponse(BaseModel):
-    """Response from guardrails system."""
-
-    hello_rasa_user_id: Optional[str] = Field(
-        default=None,
-        description="Required. User identifier for the Hello Rasa project. ",
-    )
-    hello_rasa_project_id: Optional[str] = Field(
-        default=None,
-        description="Required. Project identifier for the Hello Rasa project. ",
-    )
-    flagged: bool = Field(description="Whether any policy violations were detected.")
-    detections: List[GuardrailDetection] = Field(
-        default_factory=list, description="List of detected policy violations."
-    )
-    processing_time_ms: Optional[float] = Field(
-        default=None, description="Processing time in milliseconds."
-    )
-    metadata: Optional[Dict[str, Any]] = Field(
-        default=None, description="Additional metadata from the provider."
-    )
-
-
-class LakeraGuardrailResponse(GuardrailResponse):
-    """Response from Lakera AI `/guard` endpoint."""
-
-    @classmethod
-    def from_raw_response(
-        cls,
-        raw_response: Dict[str, Any],
-        hello_rasa_user_id: str,
-        hello_rasa_project_id: str,
-    ) -> "LakeraGuardrailResponse":
-        """Create a LakeraGuardrailResponse from a response."""
-        from rasa.builder.guardrails.utils import (
-            map_lakera_detector_type_to_guardrail_type,
-        )
-
-        # Get the basic information from the response and create the response object.
-        flagged = raw_response.get("flagged", False)
-        metadata = raw_response.get("metadata")
-        response = cls(
-            flagged=flagged,
-            metadata=metadata,
-            hello_rasa_user_id=hello_rasa_user_id,
-            hello_rasa_project_id=hello_rasa_project_id,
-        )
-
-        # If the response is not flagged, return the response object.
-        if not flagged:
-            return response
-
-        # If the response is flagged, parse the breakdown section.
-        breakdown = raw_response.get("breakdown", [])
-
-        # Parse the breakdown.
-        detections: List[GuardrailDetection] = []
-        for detector in breakdown:
-            if detector.get("detected", True):
-                detector_type = detector.get("detector_type")
-                rasa_detection_type = map_lakera_detector_type_to_guardrail_type(
-                    detector_type,
-                )
-                if not rasa_detection_type:
-                    continue
-                # Remove the detector type and the detected flag from the detector.
-                # And keep the rest as part of the metadata. In case of a Lakera this
-                # will include the message_id, project_id, policy_id, detector_id, etc.
-                metadata = copy.deepcopy(detector)
-                metadata.pop("detected")
-                metadata.pop("detector_type")
-
-                detections.append(
-                    GuardrailDetection(
-                        type=rasa_detection_type,
-                        original_type=detector_type,
-                        metadata=metadata,
-                    )
-                )
-
-        # If there are detections, add them to the response.
-        if detections:
-            response.detections = detections
-
-        return response
-
-
-class GuardrailRequestKey(BaseModel):
-    user_text: str
-    hello_rasa_user_id: str = ""
-    hello_rasa_project_id: str = ""
-    lakera_project_id: str
-
-    # hashable by value
-    model_config = ConfigDict(frozen=True)