chore: initial snapshot for gitea/github upload

llm-gateway-competitors/litellm-wheel-src/litellm/llms/openai/transcriptions/guardrail_translation/README.md

@@ -0,0 +1,159 @@
+# OpenAI Audio Transcription Guardrail Translation Handler
+
+Handler for processing OpenAI's audio transcription endpoint (`/v1/audio/transcriptions`) with guardrails.
+
+## Overview
+
+This handler processes audio transcription responses by:
+1. Applying guardrails to the transcribed text output
+2. Returning the input unchanged (since input is an audio file, not text)
+
+## Data Format
+
+### Input Format
+
+The input is an audio file, which cannot be guardrailed (it's binary data, not text).
+
+```json
+{
+  "model": "whisper-1",
+  "file": "<audio file>",
+  "response_format": "json",
+  "language": "en"
+}
+```
+
+### Output Format
+
+```json
+{
+  "text": "This is the transcribed text from the audio file."
+}
+```
+
+Or with additional metadata:
+
+```json
+{
+  "text": "This is the transcribed text from the audio file.",
+  "duration": 3.5,
+  "language": "en"
+}
+```
+
+## Usage
+
+The handler is automatically discovered and applied when guardrails are used with the audio transcription endpoint.
+
+### Example: Using Guardrails with Audio Transcription
+
+```bash
+curl -X POST 'http://localhost:4000/v1/audio/transcriptions' \
+-H 'Authorization: Bearer your-api-key' \
+-F 'file=@audio.mp3' \
+-F 'model=whisper-1' \
+-F 'guardrails=["pii_mask"]'
+```
+
+The guardrail will be applied to the **output** transcribed text only.
+
+### Example: PII Masking in Transcribed Text
+
+```bash
+curl -X POST 'http://localhost:4000/v1/audio/transcriptions' \
+-H 'Authorization: Bearer your-api-key' \
+-F 'file=@meeting_recording.mp3' \
+-F 'model=whisper-1' \
+-F 'guardrails=["mask_pii"]' \
+-F 'response_format=json'
+```
+
+If the audio contains: "My name is John Doe and my email is john@example.com"
+
+The transcription output will be: "My name is [NAME_REDACTED] and my email is [EMAIL_REDACTED]"
+
+### Example: Content Moderation on Transcriptions
+
+```bash
+curl -X POST 'http://localhost:4000/v1/audio/transcriptions' \
+-H 'Authorization: Bearer your-api-key' \
+-F 'file=@audio.wav' \
+-F 'model=whisper-1' \
+-F 'guardrails=["content_moderation"]'
+```
+
+## Implementation Details
+
+### Input Processing
+
+- **Status**: Not applicable
+- **Reason**: Input is an audio file (binary data), not text
+- **Result**: Request data returned unchanged
+
+### Output Processing
+
+- **Field**: `text` (string)
+- **Processing**: Applies guardrail to the transcribed text
+- **Result**: Updated text in response
+
+## Use Cases
+
+1. **PII Protection**: Automatically redact personally identifiable information from transcriptions
+2. **Content Filtering**: Remove or flag inappropriate content in transcribed audio
+3. **Compliance**: Ensure transcriptions meet regulatory requirements
+4. **Data Sanitization**: Clean up transcriptions before storage or further processing
+
+## Extension
+
+Override these methods to customize behavior:
+
+- `process_output_response()`: Customize how transcribed text is processed
+- `process_input_messages()`: Currently a no-op, but can be overridden if needed
+
+## Supported Call Types
+
+- `CallTypes.transcription` - Synchronous audio transcription
+- `CallTypes.atranscription` - Asynchronous audio transcription
+
+## Notes
+
+- Input processing is a no-op since audio files cannot be text-guardrailed
+- Only the transcribed text output is processed
+- Guardrails apply after transcription is complete
+- Both sync and async call types use the same handler
+- Works with all Whisper models and response formats
+
+## Common Patterns
+
+### Transcribe and Redact PII
+
+```python
+import litellm
+
+response = litellm.transcription(
+    model="whisper-1",
+    file=open("interview.mp3", "rb"),
+    guardrails=["mask_pii"],
+)
+
+# response.text will have PII redacted
+print(response.text)
+```
+
+### Async Transcription with Guardrails
+
+```python
+import litellm
+import asyncio
+
+async def transcribe_with_guardrails():
+    response = await litellm.atranscription(
+        model="whisper-1",
+        file=open("audio.mp3", "rb"),
+        guardrails=["content_filter"],
+    )
+    return response.text
+
+text = asyncio.run(transcribe_with_guardrails())
+```
+

llm-gateway-competitors/litellm-wheel-src/litellm/llms/openai/transcriptions/guardrail_translation/__init__.py

@@ -0,0 +1,13 @@
+"""OpenAI Audio Transcription handler for Unified Guardrails."""
+
+from litellm.llms.openai.transcriptions.guardrail_translation.handler import (
+    OpenAIAudioTranscriptionHandler,
+)
+from litellm.types.utils import CallTypes
+
+guardrail_translation_mappings = {
+    CallTypes.transcription: OpenAIAudioTranscriptionHandler,
+    CallTypes.atranscription: OpenAIAudioTranscriptionHandler,
+}
+
+__all__ = ["guardrail_translation_mappings", "OpenAIAudioTranscriptionHandler"]

llm-gateway-competitors/litellm-wheel-src/litellm/llms/openai/transcriptions/guardrail_translation/handler.py

@@ -0,0 +1,117 @@
+"""
+OpenAI Audio Transcription Handler for Unified Guardrails
+
+This module provides guardrail translation support for OpenAI's audio transcription endpoint.
+The handler processes the output transcribed text (input is audio, so no text to guardrail).
+"""
+
+from typing import TYPE_CHECKING, Any, Optional
+
+from litellm._logging import verbose_proxy_logger
+from litellm.llms.base_llm.guardrail_translation.base_translation import BaseTranslation
+from litellm.types.utils import GenericGuardrailAPIInputs
+
+if TYPE_CHECKING:
+    from litellm.integrations.custom_guardrail import CustomGuardrail
+    from litellm.utils import TranscriptionResponse
+
+
+class OpenAIAudioTranscriptionHandler(BaseTranslation):
+    """
+    Handler for processing OpenAI audio transcription responses with guardrails.
+
+    This class provides methods to:
+    1. Process output transcription text (post-call hook)
+
+    Note: Input processing is not applicable since the input is an audio file,
+    not text. Only the transcribed text output is processed.
+    """
+
+    async def process_input_messages(
+        self,
+        data: dict,
+        guardrail_to_apply: "CustomGuardrail",
+        litellm_logging_obj: Optional[Any] = None,
+    ) -> Any:
+        """
+        Process input - not applicable for audio transcription.
+
+        The input is an audio file, not text, so there's nothing to apply
+        guardrails to. This method returns the data unchanged.
+
+        Args:
+            data: Request data dictionary containing audio file
+            guardrail_to_apply: The guardrail instance (unused)
+
+        Returns:
+            Unmodified data (audio files don't need text guardrails)
+        """
+        verbose_proxy_logger.debug(
+            "OpenAI Audio Transcription: Input processing not applicable "
+            "(input is audio file, not text)"
+        )
+        return data
+
+    async def process_output_response(
+        self,
+        response: "TranscriptionResponse",
+        guardrail_to_apply: "CustomGuardrail",
+        litellm_logging_obj: Optional[Any] = None,
+        user_api_key_dict: Optional[Any] = None,
+    ) -> Any:
+        """
+        Process output transcription by applying guardrails to transcribed text.
+
+        Args:
+            response: Transcription response object containing transcribed text
+            guardrail_to_apply: The guardrail instance to apply
+            litellm_logging_obj: Optional logging object
+            user_api_key_dict: User API key metadata to pass to guardrails
+
+        Returns:
+            Modified response with guardrails applied to transcribed text
+        """
+        if not hasattr(response, "text") or response.text is None:
+            verbose_proxy_logger.debug(
+                "OpenAI Audio Transcription: No text in response to process"
+            )
+            return response
+
+        if isinstance(response.text, str):
+            original_text = response.text
+            # Create a request_data dict with response info and user API key metadata
+            request_data: dict = {"response": response}
+
+            # Add user API key metadata with prefixed keys
+            user_metadata = self.transform_user_api_key_dict_to_metadata(
+                user_api_key_dict
+            )
+            if user_metadata:
+                request_data["litellm_metadata"] = user_metadata
+
+            inputs = GenericGuardrailAPIInputs(texts=[original_text])
+            # Include model information from the response if available
+            if hasattr(response, "model") and response.model:
+                inputs["model"] = response.model
+            guardrailed_inputs = await guardrail_to_apply.apply_guardrail(
+                inputs=inputs,
+                request_data=request_data,
+                input_type="response",
+                logging_obj=litellm_logging_obj,
+            )
+            guardrailed_texts = guardrailed_inputs.get("texts", [])
+            response.text = guardrailed_texts[0] if guardrailed_texts else original_text
+
+            verbose_proxy_logger.debug(
+                "OpenAI Audio Transcription: Applied guardrail to transcribed text. "
+                "Original length: %d, New length: %d",
+                len(original_text),
+                len(response.text),
+            )
+        else:
+            verbose_proxy_logger.debug(
+                "OpenAI Audio Transcription: Unexpected text type: %s. Expected string.",
+                type(response.text),
+            )
+
+        return response