Add Structured Output

CHANGELOG.md

@@ -21,6 +21,8 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 - `Structure.run_stream()` for streaming Events from a Structure as an iterator.
 - Support for `GenericMessageContent` in `AnthropicPromptDriver` and `AmazonBedrockPromptDriver`.
 - Validators to `Agent` initialization.
+- `BasePromptDriver.use_native_structured_output` for enabling or disabling structured output.
+- `BasePromptDriver.native_structured_output_strategy` for changing the structured output strategy between `native` and `tool`.
 
 ### Changed
 
@@ -29,6 +31,8 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 - `PromptTask.prompt_driver` is now serialized.
 - `PromptTask` can now do everything a `ToolkitTask` can do.
 - Loosten `numpy`s version constraint to `>=1.26.4,<3`.
+- `JsonSchemaRule`s can now take a `schema.Schema` instance. Required for using a `JsonSchemaRule` with structured output.
+- `JsonSchemaRule`s will now be used for structured output if the Prompt Driver supports it.
 
 ### Fixed
 

docs/griptape-framework/drivers/prompt-drivers.md

@@ -25,6 +25,37 @@ You can pass images to the Driver if the model supports it:
 --8<-- "docs/griptape-framework/drivers/src/prompt_drivers_images.py"
 ```
 
+## Structured Output
+
+Some LLMs provide functionality often referred to as "Structured Output". This means instructing the LLM to output data in a particular format, usually JSON. This can be useful for forcing the LLM to output in a parsable format that can be used by downstream systems.
+
+Structured output can be enabled or disabled for a Prompt Driver by setting the [use_native_structured_output](../../reference/griptape/drivers.md#griptape.drivers.BasePromptDriver.use_native_structured_output).
+
+If `use_native_structured_output=True`, you can change _how_ the output is structured by setting the [native_structured_output_strategy](../../reference/griptape/drivers.md#griptape.drivers.BasePromptDriver.native_structured_output_strategy) to one of:
+
+- `native`: The Driver will use the LLM's structured output functionality provided by the API.
+- `tool`: Griptape will pass a special Tool, [StructuredOutputTool](../../reference/griptape/tools/structured_output_tool.md) and try to force the LLM to use a Tool.
+
+### JSON Schema
+
+The easiest way to get started with structured output is by using a [JsonSchemaRule](../structures/rulesets.md#json-schema). If a [schema.Schema](https://pypi.org/project/schema/) instance is provided to the Rule, Griptape will convert it to a JSON Schema and provide it to the LLM using the selected structured output strategy.
+
+```python
+--8<-- "docs/griptape-framework/drivers/src/prompt_drivers_structured_output.py"
+```
+
+### Multiple Schemas
+
+If multiple `JsonSchemaRule`s are provided, Griptape will merge them into a single JSON Schema using `anyOf`.
+
+Some LLMs may not support `anyOf` as a top-level JSON Schema. To work around this, you can try using another `native_structured_output_strategy`:
+
+```python
+--8<-- "docs/griptape-framework/drivers/src/prompt_drivers_structured_output_multi.py"
+```
+
+Not every LLM supports `use_native_structured_output` or all `native_structured_output_strategy` options.
+
 ## Prompt Drivers
 
 Griptape offers the following Prompt Drivers for interacting with LLMs.

docs/griptape-framework/drivers/src/prompt_drivers_structured_output.py

@@ -0,0 +1,35 @@
+import schema
+from rich.pretty import pprint
+
+from griptape.drivers import OpenAiChatPromptDriver
+from griptape.rules import JsonSchemaRule, Rule
+from griptape.structures import Pipeline
+from griptape.tasks import PromptTask
+
+pipeline = Pipeline(
+    tasks=[
+        PromptTask(
+            prompt_driver=OpenAiChatPromptDriver(
+                model="gpt-4o",
+                use_native_structured_output=True,
+                native_structured_output_strategy="native",
+            ),
+            rules=[
+                Rule("You are a helpful math tutor. Guide the user through the solution step by step."),
+                JsonSchemaRule(
+                    schema.Schema(
+                        {
+                            "steps": [schema.Schema({"explanation": str, "output": str})],
+                            "final_answer": str,
+                        }
+                    )
+                ),
+            ],
+        )
+    ]
+)
+
+output = pipeline.run("How can I solve 8x + 7 = -23").output.value
+
+
+pprint(output)

docs/griptape-framework/drivers/src/prompt_drivers_structured_output_multi.py

@@ -0,0 +1,28 @@
+import schema
+from rich.pretty import pprint
+
+from griptape.drivers import OpenAiChatPromptDriver
+from griptape.rules import JsonSchemaRule
+from griptape.structures import Pipeline
+from griptape.tasks import PromptTask
+
+pipeline = Pipeline(
+    tasks=[
+        PromptTask(
+            prompt_driver=OpenAiChatPromptDriver(
+                model="gpt-4o",
+                use_native_structured_output=True,
+                native_structured_output_strategy="tool",
+            ),
+            rules=[
+                JsonSchemaRule(schema.Schema({"color": "red"})),
+                JsonSchemaRule(schema.Schema({"color": "blue"})),
+            ],
+        )
+    ]
+)
+
+output = pipeline.run("Pick a color").output.value
+
+
+pprint(output)

docs/griptape-framework/misc/events.md

@@ -79,7 +79,8 @@ Handler 2 <class 'griptape.events.finish_structure_run_event.FinishStructureRunE
 You can use `Structure.run_stream()` for streaming Events from the `Structure` in the form of an iterator.
 
 !!! tip
-  Set `stream=True` on your [Prompt Driver](../drivers/prompt-drivers.md) in order to receive completion chunk events.
+
+Set `stream=True` on your [Prompt Driver](../drivers/prompt-drivers.md) in order to receive completion chunk events.
 
 ```python
 --8<-- "docs/griptape-framework/misc/src/events_streaming.py"

docs/griptape-framework/structures/rulesets.md

@@ -29,6 +29,9 @@ A [Ruleset](../../reference/griptape/rules/ruleset.md) can be used to define [Ru
 [JsonSchemaRule](../../reference/griptape/rules/json_schema_rule.md)s defines a structured format for the LLM's output by providing a JSON schema.
 This is particularly useful when you need the LLM to return well-formed data, such as JSON objects, with specific fields and data types.
 
+If the Prompt Driver supports [Structured Output](../drivers/prompt-drivers.md#structured-output), Griptape will use the schema provided to the `JsonSchemaRule` to ensure JSON output.
+If the Prompt Driver does not support Structured Output, Griptape will include the schema in the system prompt using [this template](https://github.com/griptape-ai/griptape/blob/main/griptape/templates/rules/json_schema.j2).
+
 ```python
 --8<-- "docs/griptape-framework/structures/src/json_schema_rule.py"
 ```
@@ -47,6 +50,10 @@ Although Griptape leverages the `schema` library, you're free to use any JSON sc
 
 For example, using `pydantic`:
 
+!!! warning
+
+Griptape does not yet support using `pydantic` schemas for[Structured Output](../drivers/prompt-drivers.md#structured-output). It is recommended to pass a `schema.Schema` instance.
+
 ```python
 --8<-- "docs/griptape-framework/structures/src/json_schema_rule_pydantic.py"
 ```

docs/griptape-framework/structures/src/json_schema_rule.py

@@ -5,13 +5,7 @@
 from griptape.rules.json_schema_rule import JsonSchemaRule
 from griptape.structures import Agent
 
-agent = Agent(
-    rules=[
-        JsonSchemaRule(
-            schema.Schema({"answer": str, "relevant_emojis": schema.Schema(["str"])}).json_schema("Output Format")
-        )
-    ]
-)
+agent = Agent(rules=[JsonSchemaRule(schema.Schema({"answer": str, "relevant_emojis": schema.Schema(["str"])}))])
 
 output = agent.run("What is the sentiment of this message?: 'I am so happy!'").output
 

griptape/common/prompt_stack/prompt_stack.py

@@ -1,6 +1,6 @@
 from __future__ import annotations
 
-from typing import TYPE_CHECKING
+from typing import TYPE_CHECKING, Any, Optional
 
 from attrs import define, field
 
@@ -31,6 +31,7 @@
 class PromptStack(SerializableMixin):
     messages: list[Message] = field(factory=list, kw_only=True, metadata={"serializable": True})
     tools: list[BaseTool] = field(factory=list, kw_only=True)
+    output_schema: Optional[Any] = field(default=None, kw_only=True)
 
     @property
     def system_messages(self) -> list[Message]:

griptape/drivers/__init__.py

@@ -1,3 +1,7 @@
+from .schema.base_schema_driver import BaseSchemaDriver
+from .schema.schema_schema_driver import SchemaSchemaDriver
+from .schema.pydantic_schema_driver import PydanticSchemaDriver
+
 from .prompt.base_prompt_driver import BasePromptDriver
 from .prompt.openai_chat_prompt_driver import OpenAiChatPromptDriver
 from .prompt.azure_openai_chat_prompt_driver import AzureOpenAiChatPromptDriver
@@ -240,4 +244,7 @@
     "BaseAssistantDriver",
     "GriptapeCloudAssistantDriver",
     "OpenAiAssistantDriver",
+    "BaseSchemaDriver",
+    "SchemaSchemaDriver",
+    "PydanticSchemaDriver",
 ]

griptape/drivers/prompt/amazon_bedrock_prompt_driver.py

@@ -1,9 +1,9 @@
 from __future__ import annotations
 
 import logging
-from typing import TYPE_CHECKING, Any
+from typing import TYPE_CHECKING, Any, Literal
 
-from attrs import Factory, define, field
+from attrs import Attribute, Factory, define, field
 from schema import Schema
 
 from griptape.artifacts import (
@@ -55,9 +55,20 @@ class AmazonBedrockPromptDriver(BasePromptDriver):
         kw_only=True,
     )
     use_native_tools: bool = field(default=True, kw_only=True, metadata={"serializable": True})
+    use_native_structured_output: bool = field(default=True, kw_only=True, metadata={"serializable": True})
+    native_structured_output_strategy: Literal["native", "tool"] = field(
+        default="tool", kw_only=True, metadata={"serializable": True}
+    )
     tool_choice: dict = field(default=Factory(lambda: {"auto": {}}), kw_only=True, metadata={"serializable": True})
     _client: Any = field(default=None, kw_only=True, alias="client", metadata={"serializable": False})
 
+    @native_structured_output_strategy.validator  # pyright: ignore[reportAttributeAccessIssue, reportOptionalMemberAccess]
+    def validate_native_structured_output_strategy(self, attribute: Attribute, value: str) -> str:
+        if value == "native":
+            raise ValueError("AmazonBedrockPromptDriver does not support `native` structured output mode.")
+
+        return value
+
     @lazy_property()
     def client(self) -> Any:
         return self.session.client("bedrock-runtime")
@@ -103,10 +114,9 @@ def try_stream(self, prompt_stack: PromptStack) -> Iterator[DeltaMessage]:
 
     def _base_params(self, prompt_stack: PromptStack) -> dict:
         system_messages = [{"text": message.to_text()} for message in prompt_stack.system_messages]
-
         messages = self.__to_bedrock_messages([message for message in prompt_stack.messages if not message.is_system()])
 
-        return {
+        params = {
             "modelId": self.model,
             "messages": messages,
             "system": system_messages,
@@ -115,14 +125,27 @@ def _base_params(self, prompt_stack: PromptStack) -> dict:
                 **({"maxTokens": self.max_tokens} if self.max_tokens is not None else {}),
             },
             "additionalModelRequestFields": self.additional_model_request_fields,
-            **(
-                {"toolConfig": {"tools": self.__to_bedrock_tools(prompt_stack.tools), "toolChoice": self.tool_choice}}
-                if prompt_stack.tools and self.use_native_tools
-                else {}
-            ),
             **self.extra_params,
         }
 
+        if prompt_stack.tools and self.use_native_tools:
+            params["toolConfig"] = {
+                "tools": [],
+                "toolChoice": self.tool_choice,
+            }
+
+            if (
+                prompt_stack.output_schema is not None
+                and self.use_native_structured_output
+                and self.native_structured_output_strategy == "tool"
+            ):
+                self._add_structured_output_tool(prompt_stack)
+                params["toolConfig"]["toolChoice"] = {"any": {}}
+
+            params["toolConfig"]["tools"] = self.__to_bedrock_tools(prompt_stack.tools)
+
+        return params
+
     def __to_bedrock_messages(self, messages: list[Message]) -> list[dict]:
         return [
             {
@@ -145,9 +168,7 @@ def __to_bedrock_tools(self, tools: list[BaseTool]) -> list[dict]:
                     "name": tool.to_native_tool_name(activity),
                     "description": tool.activity_description(activity),
                     "inputSchema": {
-                        "json": (tool.activity_schema(activity) or Schema({})).json_schema(
-                            "http://json-schema.org/draft-07/schema#",
-                        ),
+                        "json": self.schema_driver.to_json_schema(tool.activity_schema(activity) or Schema({})),
                     },
                 },
             }

griptape/drivers/prompt/anthropic_prompt_driver.py

@@ -1,9 +1,9 @@
 from __future__ import annotations
 
 import logging
-from typing import TYPE_CHECKING, Optional
+from typing import TYPE_CHECKING, Literal, Optional
 
-from attrs import Factory, define, field
+from attrs import Attribute, Factory, define, field
 from schema import Schema
 
 from griptape.artifacts import (
@@ -68,13 +68,24 @@ class AnthropicPromptDriver(BasePromptDriver):
     top_k: int = field(default=250, kw_only=True, metadata={"serializable": True})
     tool_choice: dict = field(default=Factory(lambda: {"type": "auto"}), kw_only=True, metadata={"serializable": False})
     use_native_tools: bool = field(default=True, kw_only=True, metadata={"serializable": True})
+    use_native_structured_output: bool = field(default=True, kw_only=True, metadata={"serializable": True})
+    native_structured_output_strategy: Literal["native", "tool"] = field(
+        default="tool", kw_only=True, metadata={"serializable": True}
+    )
     max_tokens: int = field(default=1000, kw_only=True, metadata={"serializable": True})
     _client: Client = field(default=None, kw_only=True, alias="client", metadata={"serializable": False})
 
     @lazy_property()
     def client(self) -> Client:
         return import_optional_dependency("anthropic").Anthropic(api_key=self.api_key)
 
+    @native_structured_output_strategy.validator  # pyright: ignore[reportAttributeAccessIssue, reportOptionalMemberAccess]
+    def validate_native_structured_output_strategy(self, attribute: Attribute, value: str) -> str:
+        if value == "native":
+            raise ValueError("AnthropicPromptDriver does not support `native` structured output mode.")
+
+        return value
+
     @observable
     def try_run(self, prompt_stack: PromptStack) -> Message:
         params = self._base_params(prompt_stack)
@@ -110,23 +121,33 @@ def _base_params(self, prompt_stack: PromptStack) -> dict:
         system_messages = prompt_stack.system_messages
         system_message = system_messages[0].to_text() if system_messages else None
 
-        return {
+        params = {
             "model": self.model,
             "temperature": self.temperature,
             "stop_sequences": self.tokenizer.stop_sequences,
             "top_p": self.top_p,
             "top_k": self.top_k,
             "max_tokens": self.max_tokens,
             "messages": messages,
-            **(
-                {"tools": self.__to_anthropic_tools(prompt_stack.tools), "tool_choice": self.tool_choice}
-                if prompt_stack.tools and self.use_native_tools
-                else {}
-            ),
             **({"system": system_message} if system_message else {}),
             **self.extra_params,
         }
 
+        if prompt_stack.tools and self.use_native_tools:
+            params["tool_choice"] = self.tool_choice
+
+            if (
+                prompt_stack.output_schema is not None
+                and self.use_native_structured_output
+                and self.native_structured_output_strategy == "tool"
+            ):
+                self._add_structured_output_tool(prompt_stack)
+                params["tool_choice"] = {"type": "any"}
+
+            params["tools"] = self.__to_anthropic_tools(prompt_stack.tools)
+
+        return params
+
     def __to_anthropic_messages(self, messages: list[Message]) -> list[dict]:
         return [
             {"role": self.__to_anthropic_role(message), "content": self.__to_anthropic_content(message)}