pangea-sdk: remove llm_info and llm_input from AI Guard

Author: kenany

.github/CODEOWNERS

@@ -1 +1 @@
-* @pangeacyber/sdks @doubletooth
+* @pangeacyber/sdks

CHANGELOG.md

@@ -15,6 +15,10 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 
 - Redact: `score` in `RecognizerResult` is now a float.
 
+### Removed
+
+- AI Guard: `llm_info` and `llm_input`.
+
 ## 5.5.1 - 2025-02-17
 
 ### Changed

packages/pangea-sdk/pangea/asyncio/services/ai_guard.py

@@ -58,7 +58,6 @@ async def guard_text(
         *,
         recipe: str | None = None,
         debug: bool | None = None,
-        llm_info: str | None = None,
         log_fields: LogFields | None = None,
     ) -> PangeaResponse[TextGuardResult[None]]:
         """
@@ -78,7 +77,6 @@ async def guard_text(
                 are to be applied to the text, such as defang malicious URLs.
             debug: Setting this value to true will provide a detailed analysis
                 of the text data
-            llm_info: Short string hint for the LLM Provider information
             log_field: Additional fields to include in activity log
 
         Examples:
@@ -92,7 +90,6 @@ async def guard_text(
         messages: _T,
         recipe: str | None = None,
         debug: bool | None = None,
-        llm_info: str | None = None,
         log_fields: LogFields | None = None,
     ) -> PangeaResponse[TextGuardResult[_T]]:
         """
@@ -113,59 +110,19 @@ async def guard_text(
                 are to be applied to the text, such as defang malicious URLs.
             debug: Setting this value to true will provide a detailed analysis
                 of the text data
-            llm_info: Short string hint for the LLM Provider information
             log_field: Additional fields to include in activity log
 
         Examples:
             response = await ai_guard.guard_text(messages=[{"role": "user", "content": "hello world"}])
         """
 
-    @overload
-    async def guard_text(
-        self,
-        *,
-        llm_input: _T,
-        recipe: str | None = None,
-        debug: bool | None = None,
-        llm_info: str | None = None,
-        log_fields: LogFields | None = None,
-    ) -> PangeaResponse[TextGuardResult[_T]]:
-        """
-        Text Guard for scanning LLM inputs and outputs
-
-        Analyze and redact text to avoid manipulation of the model, addition of
-        malicious content, and other undesirable data transfers.
-
-        OperationId: ai_guard_post_v1_text_guard
-
-        Args:
-            llm_input: Structured full llm payload data to be scanned by AI
-                Guard for PII, sensitive data, malicious content, and other data
-                types defined by the configuration. Supports processing up to
-                10KB of JSON text
-            recipe: Recipe key of a configuration of data types and settings
-                defined in the Pangea User Console. It specifies the rules that
-                are to be applied to the text, such as defang malicious URLs.
-            debug: Setting this value to true will provide a detailed analysis
-                of the text data
-            llm_info: Short string hint for the LLM Provider information
-            log_field: Additional fields to include in activity log
-
-        Examples:
-            response = await ai_guard.guard_text(
-                llm_input={"model": "gpt-4o", "messages": [{"role": "user", "content": "hello world"}]}
-            )
-        """
-
     async def guard_text(  # type: ignore[misc]
         self,
         text: str | None = None,
         *,
         messages: _T | None = None,
-        llm_input: _T | None = None,
         recipe: str | None = None,
         debug: bool | None = None,
-        llm_info: str | None = None,
         log_fields: LogFields | None = None,
     ) -> PangeaResponse[TextGuardResult[None]]:
         """
@@ -184,38 +141,28 @@ async def guard_text(  # type: ignore[misc]
                 PII, sensitive data, malicious content, and other data types
                 defined by the configuration. Supports processing up to 10KB of
                 JSON text
-            llm_input: Structured full llm payload data to be scanned by AI
-                Guard for PII, sensitive data, malicious content, and other data
-                types defined by the configuration. Supports processing up to
-                10KB of JSON text
             recipe: Recipe key of a configuration of data types and settings
                 defined in the Pangea User Console. It specifies the rules that
                 are to be applied to the text, such as defang malicious URLs.
             debug: Setting this value to true will provide a detailed analysis
                 of the text data
-            llm_info: Short string hint for the LLM Provider information
             log_field: Additional fields to include in activity log
 
         Examples:
             response = await ai_guard.guard_text("text")
         """
 
-        if not any((text, messages, llm_input)):
-            raise ValueError("Exactly one of `text`, `messages`, or `llm_input` must be given")
-
-        if sum((text is not None, messages is not None, llm_input is not None)) > 1:
-            raise ValueError("Only one of `text`, `messages`, or `llm_input` can be given at once")
+        if text is not None and messages is not None:
+            raise ValueError("Exactly one of `text` or `messages` must be given")
 
         return await self.request.post(
             "v1/text/guard",
             TextGuardResult,
             data={
                 "text": text,
                 "messages": messages,
-                "llm_input": llm_input,
                 "recipe": recipe,
                 "debug": debug,
-                "llm_info": llm_info,
                 "log_fields": log_fields,
             },
         )

packages/pangea-sdk/pangea/services/ai_guard.py

@@ -160,7 +160,6 @@ def guard_text(
         *,
         recipe: str | None = None,
         debug: bool | None = None,
-        llm_info: str | None = None,
         log_fields: LogFields | None = None,
     ) -> PangeaResponse[TextGuardResult[None]]:
         """
@@ -180,7 +179,6 @@ def guard_text(
                 are to be applied to the text, such as defang malicious URLs.
             debug: Setting this value to true will provide a detailed analysis
                 of the text data
-            llm_info: Short string hint for the LLM Provider information
             log_field: Additional fields to include in activity log
 
         Examples:
@@ -194,7 +192,6 @@ def guard_text(
         messages: _T,
         recipe: str | None = None,
         debug: bool | None = None,
-        llm_info: str | None = None,
         log_fields: LogFields | None = None,
     ) -> PangeaResponse[TextGuardResult[_T]]:
         """
@@ -215,59 +212,19 @@ def guard_text(
                 are to be applied to the text, such as defang malicious URLs.
             debug: Setting this value to true will provide a detailed analysis
                 of the text data
-            llm_info: Short string hint for the LLM Provider information
             log_field: Additional fields to include in activity log
 
         Examples:
             response = ai_guard.guard_text(messages=[{"role": "user", "content": "hello world"}])
         """
 
-    @overload
-    def guard_text(
-        self,
-        *,
-        llm_input: _T,
-        recipe: str | None = None,
-        debug: bool | None = None,
-        llm_info: str | None = None,
-        log_fields: LogFields | None = None,
-    ) -> PangeaResponse[TextGuardResult[_T]]:
-        """
-        Text Guard for scanning LLM inputs and outputs
-
-        Analyze and redact text to avoid manipulation of the model, addition of
-        malicious content, and other undesirable data transfers.
-
-        OperationId: ai_guard_post_v1_text_guard
-
-        Args:
-            llm_input: Structured full llm payload data to be scanned by AI
-                Guard for PII, sensitive data, malicious content, and other data
-                types defined by the configuration. Supports processing up to
-                10KB of JSON text
-            recipe: Recipe key of a configuration of data types and settings
-                defined in the Pangea User Console. It specifies the rules that
-                are to be applied to the text, such as defang malicious URLs.
-            debug: Setting this value to true will provide a detailed analysis
-                of the text data
-            llm_info: Short string hint for the LLM Provider information
-            log_field: Additional fields to include in activity log
-
-        Examples:
-            response = ai_guard.guard_text(
-                llm_input={"model": "gpt-4o", "messages": [{"role": "user", "content": "hello world"}]}
-            )
-        """
-
     def guard_text(  # type: ignore[misc]
         self,
         text: str | None = None,
         *,
         messages: _T | None = None,
-        llm_input: _T | None = None,
         recipe: str | None = None,
         debug: bool | None = None,
-        llm_info: str | None = None,
         log_fields: LogFields | None = None,
     ) -> PangeaResponse[TextGuardResult[None]]:
         """
@@ -286,38 +243,28 @@ def guard_text(  # type: ignore[misc]
                 PII, sensitive data, malicious content, and other data types
                 defined by the configuration. Supports processing up to 10KB of
                 JSON text
-            llm_input: Structured full llm payload data to be scanned by AI
-                Guard for PII, sensitive data, malicious content, and other data
-                types defined by the configuration. Supports processing up to
-                10KB of JSON text
             recipe: Recipe key of a configuration of data types and settings
                 defined in the Pangea User Console. It specifies the rules that
                 are to be applied to the text, such as defang malicious URLs.
             debug: Setting this value to true will provide a detailed analysis
                 of the text data
-            llm_info: Short string hint for the LLM Provider information
             log_field: Additional fields to include in activity log
 
         Examples:
             response = ai_guard.guard_text("text")
         """
 
-        if not any((text, messages, llm_input)):
-            raise ValueError("At least one of `text`, `messages`, or `llm_input` must be given")
-
-        if sum((text is not None, messages is not None, llm_input is not None)) > 1:
-            raise ValueError("Only one of `text`, `messages`, or `llm_input` can be given at once")
+        if text is not None and messages is not None:
+            raise ValueError("Exactly one of `text` or `messages` must be given")
 
         return self.request.post(
             "v1/text/guard",
             TextGuardResult,
             data={
                 "text": text,
                 "messages": messages,
-                "llm_input": llm_input,
                 "recipe": recipe,
                 "debug": debug,
-                "llm_info": llm_info,
                 "log_fields": log_fields,
             },
         )

packages/pangea-sdk/tests/integration/asyncio/test_ai_guard.py

@@ -44,11 +44,3 @@ async def test_text_guard_messages(self) -> None:
         assert response.status == "Success"
         assert response.result
         assert response.result.prompt_messages
-
-    async def test_text_guard_llm_input(self) -> None:
-        response = await self.client.guard_text(
-            llm_input={"model": "gpt-4o", "messages": [{"role": "user", "content": "hello world"}]}
-        )
-        assert response.status == "Success"
-        assert response.result
-        assert response.result.prompt_messages

packages/pangea-sdk/tests/integration/test_ai_guard.py

@@ -37,11 +37,3 @@ def test_text_guard_messages(self) -> None:
         assert response.status == "Success"
         assert response.result
         assert response.result.prompt_messages
-
-    def test_text_guard_llm_input(self) -> None:
-        response = self.client.guard_text(
-            llm_input={"model": "gpt-4o", "messages": [{"role": "user", "content": "hello world"}]}
-        )
-        assert response.status == "Success"
-        assert response.result
-        assert response.result.prompt_messages