feat: add Pangea AI Guard community integration

docs/user-guides/community/pangea.md

@@ -0,0 +1,85 @@
+# Pangea AI Guard integration
+
+The Pangea guardrail uses configurable detection policies (called *recipes*) from the [AI Guard service](https://pangea.cloud/docs/ai-guard/) to identify and mitigate risks in AI application traffic, including:
+
+- Prompt injection attacks (with over 99% efficacy)
+- 50+ types of PII and sensitive content, with support for custom patterns
+- Toxicity, violence, self-harm, and other unwanted content
+- Malicious links, IPs, and domains
+- 100 spoken languages, with allowlist and denylist controls
+
+All detections are logged in an audit trail for analysis, attribution, and incident response.
+You can also configure webhooks to trigger alerts for specific detection types.
+
+The following environment variable is required to use the Pangea AI Guard integration:
+
+- `PANGEA_API_TOKEN`: Pangea API token with access to the AI Guard service.
+
+You can also optionally set:
+
+- `PANGEA_BASE_URL_TEMPLATE`: Template for constructing the base URL for API requests. The `{SERVICE_NAME}` placeholder will be replaced with the service name slug.
+  Defaults to `https://ai-guard.aws.us.pangea.cloud` for Pangea's hosted (SaaS) deployment.
+
+## Setup
+
+Colang v1:
+
+```yaml
+# config.yml
+
+rails:
+  config:
+    pangea:
+      input:
+        recipe: pangea_prompt_guard
+      output:
+        recipe: pangea_llm_response_guard
+
+  input:
+    flows:
+      - pangea ai guard input
+
+  output:
+    flows:
+      - pangea ai guard output
+```
+
+Colang v2:
+
+```yaml
+# config.yml
+
+colang_version: "2.x"
+
+rails:
+  config:
+    pangea:
+      input:
+        recipe: pangea_prompt_guard
+      output:
+        recipe: pangea_llm_response_guard
+```
+
+```
+# rails.co
+
+import guardrails
+import nemoguardrails.library.pangea
+
+flow input rails $input_text
+    pangea ai guard input
+
+flow output rails $output_text
+    pangea ai guard output
+```
+
+## Next steps
+
+- Explore example configurations for integrating Pangea AI Guard with your preferred Colang version:
+  - [Pangea AI Guard for NeMo Guardrails v1](https://github.com/NVIDIA/NeMo-Guardrails/tree/develop/examples/configs/pangea)
+  - [Pangea AI Guard for NeMo Guardrails v2](https://github.com/NVIDIA/NeMo-Guardrails/tree/develop/examples/configs/pangea_v2)
+  - [Pangea AI Guard without LLM (guardrails only)](https://github.com/NVIDIA/NeMo-Guardrails/tree/develop/examples/configs/pangea_v2_no_llm) – Use this setup to evaluate AI Guard’s detection and response capabilities independently.
+- Adjust your detection policies to fit your application’s risk profile. See the [AI Guard Recipes](https://pangea.cloud/docs/ai-guard/recipes) documentation for configuration details.
+- Enable [AI Guard webhooks](https://pangea.cloud/docs/ai-guard/recipes#add-webhooks-to-detectors) to receive real-time alerts for detections in your NeMo Guardrails-powered application.
+- Monitor and analyze detection activity in the [AI Guard Activity Log](https://pangea.cloud/docs/ai-guard/activity-log) for auditing and attribution.
+- Learn more about [AI Guard Deployment Options](https://pangea.cloud/docs/deployment-models/) to understand how and where AI Guard can run to protect your AI applications.

docs/user-guides/guardrails-library.md

@@ -25,6 +25,7 @@ NeMo Guardrails comes with a library of built-in guardrails that you can easily
    - [Private AI PII detection](#private-ai-pii-detection)
    - [Fiddler Guardrails for Safety and Hallucination Detection](#fiddler-guardrails-for-safety-and-hallucination-detection)
    - [Prompt Security Protection](#prompt-security-protection)
+   - [Pangea AI Guard](#pangea-ai-guard)
    - OpenAI Moderation API - *[COMING SOON]*
 
 4. Other
@@ -866,6 +867,26 @@ rails:
 
 For more details, check out the [Prompt Security Integration](./community/prompt-security.md) page.
 
+### Pangea AI Guard
+
+NeMo Guardrails supports using [Pangea AI Guard](https://pangea.cloud/services/ai-guard/) for protecting data and
+interactions with LLMs within AI-powered applications.
+
+#### Example usage
+
+```yaml
+rails:
+  input:
+    flows:
+      - pangea ai guard input
+
+  output:
+    flows:
+      - pangea ai guard output
+```
+
+For more details, check out the [Pangea AI Guard Integration](./community/pangea.md) page.
+
 ## Other
 
 ### Jailbreak Detection

docs/user-guides/llm-support.md

@@ -40,6 +40,7 @@ If you want to use an LLM and you cannot see a prompt in the [prompts folder](ht
 | Patronus Evaluate API _(LLM independent)_          | ✔                         | ✔                         | ✔                         | ✔                         | ✔                         | ✔                  | ✔                    | ✔                    | ✔                    | ✔                    | ✔                                 |
 | Fiddler Fast Faitfhulness Hallucination Detection _(LLM independent)_          | ✔                         | ✔                         | ✔                         | ✔                         | ✔                         | ✔                  | ✔                    | ✔                    | ✔                    | ✔                    | ✔
 | Fiddler Fast Safety & Jailbreak Detection _(LLM independent)_          | ✔                         | ✔                         | ✔                         | ✔                         | ✔                         | ✔                  | ✔                    | ✔                    | ✔                    | ✔                    | ✔                     |
+| Pangea AI Guard integration _(LLM independent)_          | ✔                         | ✔                         | ✔                         | ✔                         | ✔                         | ✔                  | ✔                    | ✔                    | ✔                    | ✔                    | ✔                     |
 
 Table legend:
 

examples/configs/pangea/README.md

@@ -0,0 +1,14 @@
+# Pangea Example
+
+This example demonstrates how to integrate with the [Pangea AI Guard](https://pangea.cloud/services/ai-guard/) API for protecting data and interactions with LLMs within AI-powered applications
+
+To test this configuration you can use the CLI Chat by running the following command from the `examples/configs/pangea` directory:
+
+```bash
+poetry run nemoguardrails chat --config=.
+```
+
+Documentation:
+
+- [Full Pangea integration guide](../../../docs/user-guides/community/pangea.md)
+- [Configuration options and setup instructions](../../../docs/user-guides/community/pangea.md#setup)

examples/configs/pangea/config.yml

@@ -0,0 +1,24 @@
+models:
+  - type: main
+    engine: openai
+    model: gpt-4o-mini
+
+instructions:
+  - type: general
+    content: |
+      You are a helpful assistant.
+
+rails:
+  config:
+    pangea:
+      input:
+        recipe: pangea_prompt_guard
+      output:
+        recipe: pangea_llm_response_guard
+
+  input:
+    flows:
+      - pangea ai guard input
+  output:
+    flows:
+      - pangea ai guard output

examples/configs/pangea_v2/README.md

@@ -0,0 +1,14 @@
+# Pangea Example
+
+This example demonstrates how to integrate with the [Pangea AI Guard](https://pangea.cloud/services/ai-guard/) API for protecting data and interactions with LLMs within AI-powered applications
+
+To test this configuration you can use the CLI Chat by running the following command from the `examples/configs/pangea_v2` directory:
+
+```bash
+poetry run nemoguardrails chat --config=.
+```
+
+Documentation:
+
+- [Full Pangea integration guide](../../../docs/user-guides/community/pangea.md)
+- [Configuration options and setup instructions](../../../docs/user-guides/community/pangea.md#setup)

examples/configs/pangea_v2/config.yml

@@ -0,0 +1,19 @@
+colang_version: "2.x"
+
+models:
+  - type: main
+    engine: openai
+    model: gpt-4o-mini
+
+instructions:
+  - type: general
+    content: |
+      You are a helpful assistant.
+
+rails:
+  config:
+    pangea:
+      input:
+        recipe: pangea_prompt_guard
+      output:
+        recipe: pangea_llm_response_guard

examples/configs/pangea_v2/main.co

@@ -0,0 +1,5 @@
+import core
+import llm
+
+flow main
+  activate llm continuation

examples/configs/pangea_v2/rails.co

@@ -0,0 +1,8 @@
+import guardrails
+import nemoguardrails.library.pangea
+
+flow input rails $input_text
+    pangea ai guard input
+
+flow output rails $output_text
+    pangea ai guard output

examples/configs/pangea_v2_no_llm/config.yml

@@ -0,0 +1,12 @@
+colang_version: "2.x"
+
+# No models section - guardrails only mode
+# No LLM is required since we're only using Pangea APIs
+
+rails:
+  config:
+    pangea:
+      input:
+        recipe: pangea_prompt_guard
+      output:
+        recipe: pangea_llm_response_guard

examples/configs/pangea_v2_no_llm/main.co

@@ -0,0 +1,12 @@
+import core
+
+flow main
+    activate message handler
+
+# Allow continuation after blocked messages in guardrails only mode
+flow message handler
+    when user said something
+        global $user_message
+        # At this point, $user_message contains the processed value from input rails
+        bot say "Processed message: {$user_message}"
+        activate message handler  # Reactivate for next message

examples/configs/pangea_v2_no_llm/rails.co

@@ -0,0 +1,8 @@
+import guardrails
+import nemoguardrails.library.pangea
+
+flow input rails $input_text
+    pangea ai guard input
+
+flow output rails $output_text
+    pangea ai guard output

nemoguardrails/library/pangea/__init__.py

@@ -0,0 +1,14 @@
+# SPDX-FileCopyrightText: Copyright (c) 2023 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
+# SPDX-License-Identifier: Apache-2.0
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+# http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.