Add agent response format for structured output

src/llama_stack_client/_client.py

@@ -126,6 +126,7 @@ def __init__(
             if default_headers is None:
                 default_headers = {}
             default_headers["X-LlamaStack-ProviderData"] = json.dumps(provider_data)
+
         super().__init__(
             version=__version__,
             base_url=base_url,
@@ -316,6 +317,7 @@ def __init__(
             if default_headers is None:
                 default_headers = {}
             default_headers["X-LlamaStack-ProviderData"] = json.dumps(provider_data)
+
         super().__init__(
             version=__version__,
             base_url=base_url,

src/llama_stack_client/types/inference_chat_completion_params.py

@@ -18,8 +18,8 @@
     "Message",
     "Logprobs",
     "ResponseFormat",
-    "ResponseFormatJsonSchemaFormat",
-    "ResponseFormatGrammarFormat",
+    "ResponseFormatUnionMember0",
+    "ResponseFormatUnionMember1",
     "Tool",
     "InferenceChatCompletionParamsNonStreaming",
     "InferenceChatCompletionParamsStreaming",
@@ -64,19 +64,19 @@ class Logprobs(TypedDict, total=False):
     top_k: int
 
 
-class ResponseFormatJsonSchemaFormat(TypedDict, total=False):
+class ResponseFormatUnionMember0(TypedDict, total=False):
     json_schema: Required[Dict[str, Union[bool, float, str, Iterable[object], object, None]]]
 
     type: Required[Literal["json_schema"]]
 
 
-class ResponseFormatGrammarFormat(TypedDict, total=False):
+class ResponseFormatUnionMember1(TypedDict, total=False):
     bnf: Required[Dict[str, Union[bool, float, str, Iterable[object], object, None]]]
 
     type: Required[Literal["grammar"]]
 
 
-ResponseFormat: TypeAlias = Union[ResponseFormatJsonSchemaFormat, ResponseFormatGrammarFormat]
+ResponseFormat: TypeAlias = Union[ResponseFormatUnionMember0, ResponseFormatUnionMember1]
 
 
 class Tool(TypedDict, total=False):

src/llama_stack_client/types/inference_completion_params.py

@@ -13,8 +13,8 @@
     "InferenceCompletionParamsBase",
     "Logprobs",
     "ResponseFormat",
-    "ResponseFormatJsonSchemaFormat",
-    "ResponseFormatGrammarFormat",
+    "ResponseFormatUnionMember0",
+    "ResponseFormatUnionMember1",
     "InferenceCompletionParamsNonStreaming",
     "InferenceCompletionParamsStreaming",
 ]
@@ -38,19 +38,19 @@ class Logprobs(TypedDict, total=False):
     top_k: int
 
 
-class ResponseFormatJsonSchemaFormat(TypedDict, total=False):
+class ResponseFormatUnionMember0(TypedDict, total=False):
     json_schema: Required[Dict[str, Union[bool, float, str, Iterable[object], object, None]]]
 
     type: Required[Literal["json_schema"]]
 
 
-class ResponseFormatGrammarFormat(TypedDict, total=False):
+class ResponseFormatUnionMember1(TypedDict, total=False):
     bnf: Required[Dict[str, Union[bool, float, str, Iterable[object], object, None]]]
 
     type: Required[Literal["grammar"]]
 
 
-ResponseFormat: TypeAlias = Union[ResponseFormatJsonSchemaFormat, ResponseFormatGrammarFormat]
+ResponseFormat: TypeAlias = Union[ResponseFormatUnionMember0, ResponseFormatUnionMember1]
 
 
 class InferenceCompletionParamsNonStreaming(InferenceCompletionParamsBase, total=False):

src/llama_stack_client/types/shared/agent_config.py

@@ -1,6 +1,6 @@
 # File generated from our OpenAPI spec by Stainless. See CONTRIBUTING.md for details.
 
-from typing import List, Union, Optional
+from typing import Dict, List, Union, Optional
 from typing_extensions import Literal, TypeAlias
 
 from ..._models import BaseModel
@@ -12,7 +12,22 @@
 from .wolfram_alpha_tool_definition import WolframAlphaToolDefinition
 from .code_interpreter_tool_definition import CodeInterpreterToolDefinition
 
-__all__ = ["AgentConfig", "Tool"]
+__all__ = ["AgentConfig", "ResponseFormat", "ResponseFormatUnionMember0", "ResponseFormatUnionMember1", "Tool"]
+
+
+class ResponseFormatUnionMember0(BaseModel):
+    json_schema: Dict[str, Union[bool, float, str, List[object], object, None]]
+
+    type: Literal["json_schema"]
+
+
+class ResponseFormatUnionMember1(BaseModel):
+    bnf: Dict[str, Union[bool, float, str, List[object], object, None]]
+
+    type: Literal["grammar"]
+
+
+ResponseFormat: TypeAlias = Union[ResponseFormatUnionMember0, ResponseFormatUnionMember1]
 
 Tool: TypeAlias = Union[
     SearchToolDefinition,
@@ -37,6 +52,8 @@ class AgentConfig(BaseModel):
 
     output_shields: Optional[List[str]] = None
 
+    response_format: Optional[ResponseFormat] = None
+
     sampling_params: Optional[SamplingParams] = None
 
     tool_choice: Optional[Literal["auto", "required"]] = None

src/llama_stack_client/types/shared_params/agent_config.py

@@ -2,7 +2,7 @@
 
 from __future__ import annotations
 
-from typing import List, Union, Iterable
+from typing import Dict, List, Union, Iterable
 from typing_extensions import Literal, Required, TypeAlias, TypedDict
 
 from .sampling_params import SamplingParams
@@ -13,7 +13,22 @@
 from .wolfram_alpha_tool_definition import WolframAlphaToolDefinition
 from .code_interpreter_tool_definition import CodeInterpreterToolDefinition
 
-__all__ = ["AgentConfig", "Tool"]
+__all__ = ["AgentConfig", "ResponseFormat", "ResponseFormatUnionMember0", "ResponseFormatUnionMember1", "Tool"]
+
+
+class ResponseFormatUnionMember0(TypedDict, total=False):
+    json_schema: Required[Dict[str, Union[bool, float, str, Iterable[object], object, None]]]
+
+    type: Required[Literal["json_schema"]]
+
+
+class ResponseFormatUnionMember1(TypedDict, total=False):
+    bnf: Required[Dict[str, Union[bool, float, str, Iterable[object], object, None]]]
+
+    type: Required[Literal["grammar"]]
+
+
+ResponseFormat: TypeAlias = Union[ResponseFormatUnionMember0, ResponseFormatUnionMember1]
 
 Tool: TypeAlias = Union[
     SearchToolDefinition,
@@ -38,6 +53,8 @@ class AgentConfig(TypedDict, total=False):
 
     output_shields: List[str]
 
+    response_format: ResponseFormat
+
     sampling_params: SamplingParams
 
     tool_choice: Literal["auto", "required"]

tests/api_resources/test_agents.py

@@ -39,6 +39,10 @@ def test_method_create_with_all_params(self, client: LlamaStackClient) -> None:
                 "model": "model",
                 "input_shields": ["string"],
                 "output_shields": ["string"],
+                "response_format": {
+                    "json_schema": {"foo": True},
+                    "type": "json_schema",
+                },
                 "sampling_params": {
                     "strategy": "greedy",
                     "max_tokens": 0,
@@ -169,6 +173,10 @@ async def test_method_create_with_all_params(self, async_client: AsyncLlamaStack
                 "model": "model",
                 "input_shields": ["string"],
                 "output_shields": ["string"],
+                "response_format": {
+                    "json_schema": {"foo": True},
+                    "type": "json_schema",
+                },
                 "sampling_params": {
                     "strategy": "greedy",
                     "max_tokens": 0,

tests/test_client.py

@@ -1631,7 +1631,7 @@ def test_get_platform(self) -> None:
         import threading
 
         from llama_stack_client._utils import asyncify
-        from llama_stack_client._base_client import get_platform
+        from llama_stack_client._base_client import get_platform 
 
         async def test_main() -> None:
             result = await asyncify(get_platform)()