Feature/content safety

docs/docs/integrations/tools/azure_content_safety.ipynb

@@ -0,0 +1,314 @@
+{
+ "cells": [
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "# `AzureContentSafetyTextTool`"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "## Overview"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "### Integration details"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    ">The `AzureContentSafetyTextTool` acts as a wrapper around the Azure AI Content Safety Service/API.\n",
+    ">The Tool will detect harmful content according to Azure's Content Safety Policy."
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "### Tool features"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "This integration allows for the detection of harmful or offensive content in text using Azure's Content Safety API. It supports four categories of harmful content: Sexual, Harm, Self-Harm, and Violence."
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "## Setup"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "This section provides details about how the Azure AI Content Safety integration works, including setup."
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 1,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "import getpass\n",
+    "import os"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "Now we can use the `AzureContentSafetyTextTool` combined with a model, using `create_react_agent`."
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 3,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "from langgraph.prebuilt import create_react_agent"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 7,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "from langchain_openai import AzureChatOpenAI\n",
+    "\n",
+    "from libs.community.langchain_community.tools.azure_ai_services.content_safety import (\n",
+    "    AzureContentSafetyTextTool,\n",
+    ")"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "### Credentials"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "Credentials can be set by being passed as parameters and should be stored locally as environment variables."
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 14,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "os.environ[\"CONTENT_SAFETY_ENDPOINT\"] = getpass.getpass()\n",
+    "os.environ[\"CONTENT_SAFETY_KEY\"] = getpass.getpass()"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 4,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "content_endpoint = os.environ[\"CONTENT_SAFETY_ENDPOINT\"]\n",
+    "content_key = os.environ[\"CONTENT_SAFETY_API_KEY\"]"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "## Instantiation"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "Credentials can be passed directly, but they can also be retrieved automatically by the constructor if environment variables named `CONTENT_SAFETY_ENDPOINT` and `CONTENT_SAFETY_KEY` are set."
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "os.environ[\"AZURE_OPENAI_API_KEY\"] = getpass.getpass()\n",
+    "os.environ[\"OPENAI_API_VERSION\"] = getpass.getpass()\n",
+    "os.environ[\"GPT_MODEL\"] = getpass.getpass()\n",
+    "os.environ[\"AZURE_OPENAI_ENDPOINT\"] = getpass.getpass()"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 8,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "model = AzureChatOpenAI(\n",
+    "    openai_api_version=os.environ[\"OPENAI_API_VERSION\"],\n",
+    "    azure_deployment=os.environ[\"GPT_MODEL\"],\n",
+    "    azure_endpoint=os.environ[\"AZURE_OPENAI_ENDPOINT\"],\n",
+    "    api_key=os.environ[\"AZURE_OPENAI_API_KEY\"],\n",
+    ")"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 9,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "cs = AzureContentSafetyTextTool(\n",
+    "    content_safety_key=content_key,\n",
+    "    content_safety_endpoint=content_endpoint,\n",
+    ")\n",
+    "\n",
+    "tools = [cs]"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "Create a react agent to invoke the tool. "
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 10,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "agent = create_react_agent(model, tools)"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "## Invocation"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "Input must be in the form of a string (`str`)."
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 11,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "input = \"I hate you\""
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "### [Invoke directly with args](/docs/concepts/#invoke-with-just-the-arguments)"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "Firstly, the tool can be used by directly passing input as an argument. However, this is discouraged as the tool is intended to be used in an executor chain."
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "cs.invoke({\"query\": input})"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "### [Invoke with ToolCall](/docs/concepts/#invoke-with-toolcall)"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "By using `.invoke`, the model can be told what to do and assess if using the tools it was given would assist in it's response. This is the intended use."
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "agent.invoke(\n",
+    "    {\n",
+    "        \"messages\": [\n",
+    "            (\"user\", f\"Can you check the following text for harmful content : {input}\")\n",
+    "        ]\n",
+    "    }\n",
+    ")"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "## API reference"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "[Azure AI Content Safety Overview](https://learn.microsoft.com/azure/ai-services/content-safety/overview) | [Azure AI Content Safety Python API](https://learn.microsoft.com/python/api/overview/azure/ai-contentsafety-readme?view=azure-python)"
+   ]
+  }
+ ],
+ "metadata": {
+  "kernelspec": {
+   "display_name": "Python 3",
+   "language": "python",
+   "name": "python3"
+  },
+  "language_info": {
+   "codemirror_mode": {
+    "name": "ipython",
+    "version": 3
+   },
+   "file_extension": ".py",
+   "mimetype": "text/x-python",
+   "name": "python",
+   "nbconvert_exporter": "python",
+   "pygments_lexer": "ipython3",
+   "version": "3.12.8"
+  }
+ },
+ "nbformat": 4,
+ "nbformat_minor": 2
+}

libs/community/extended_testing_deps.txt

@@ -4,6 +4,7 @@ anthropic>=0.3.11,<0.4
 arxiv>=1.4,<2
 assemblyai>=0.17.0,<0.18
 atlassian-python-api>=3.36.0,<4
+azure-ai-contentsafety>=1.0.0
 azure-ai-documentintelligence>=1.0.0b1,<2
 azure-identity>=1.15.0,<2
 azure-search-documents==11.4.0