Home-Assistant/core
1
0
Fork 0
mirror of https://github.com/home-assistant/core.git synced 2026-02-15 07:36:16 +00:00
Code Activity

Anthropic Structured Outputs support (#162515)

Browse Source
This commit is contained in:
Denis Shulyaka
2026-02-12 01:25:46 +03:00
committed by GitHub
parent 661b14dec5
commit e060395786
6 changed files with 115 additions and 38 deletions
Download Patch File Download Diff File Expand all files Collapse all files

22
homeassistant/components/anthropic/config_flow.py
View File

@@ -491,22 +491,24 @@ class ConversationSubentryFlowHandler(ConfigSubentryFlow):
"role": "user",
"content": "Where are the following coordinates located: "
f"({zone_home.attributes[ATTR_LATITUDE]},"
f" {zone_home.attributes[ATTR_LONGITUDE]})? Please respond "
"only with a JSON object using the following schema:\n"
f"{convert(location_schema)}",
},
{
"role": "assistant",
"content": "{", # hints the model to skip any preamble
},
f" {zone_home.attributes[ATTR_LONGITUDE]})?",
}
],
max_tokens=cast(int, DEFAULT[CONF_MAX_TOKENS]),
output_config={
"format": {
"type": "json_schema",
"schema": {
**convert(location_schema),
"additionalProperties": False,
},
}
},
)
_LOGGER.debug("Model response: %s", response.content)
location_data = location_schema(
json.loads(
"{"
+ "".join(
"".join(
block.text
for block in response.content
if isinstance(block, anthropic.types.TextBlock)

9
homeassistant/components/anthropic/const.py
View File

@@ -56,6 +56,15 @@ NON_ADAPTIVE_THINKING_MODELS = [
"claude-3",
]
UNSUPPORTED_STRUCTURED_OUTPUT_MODELS = [
"claude-opus-4-1",
"claude-opus-4-0",
"claude-opus-4-20250514",
"claude-sonnet-4-0",
"claude-sonnet-4-20250514",
"claude-3",
]
WEB_SEARCH_UNSUPPORTED_MODELS = [
"claude-3-haiku",
"claude-3-opus",

52
homeassistant/components/anthropic/entity.py
View File

@@ -20,6 +20,7 @@ from anthropic.types import (
DocumentBlockParam,
ImageBlockParam,
InputJSONDelta,
JSONOutputFormatParam,
MessageDeltaUsage,
MessageParam,
MessageStreamEvent,
@@ -94,6 +95,7 @@ from .const import (
MIN_THINKING_BUDGET,
NON_ADAPTIVE_THINKING_MODELS,
NON_THINKING_MODELS,
UNSUPPORTED_STRUCTURED_OUTPUT_MODELS,
)
# Max number of back and forth with the LLM to generate a response
@@ -697,8 +699,25 @@ class AnthropicBaseLLMEntity(Entity):
)
if structure and structure_name:
structure_name = slugify(structure_name)
if model_args["thinking"]["type"] == "disabled":
if not model.startswith(tuple(UNSUPPORTED_STRUCTURED_OUTPUT_MODELS)):
# Native structured output for those models who support it.
structure_name = None
model_args.setdefault("output_config", OutputConfigParam())[
"format"
] = JSONOutputFormatParam(
type="json_schema",
schema={
**convert(
structure,
custom_serializer=chat_log.llm_api.custom_serializer
if chat_log.llm_api
else llm.selector_serializer,
),
"additionalProperties": False,
},
)
elif model_args["thinking"]["type"] == "disabled":
structure_name = slugify(structure_name)
if not tools:
# Simplest case: no tools and no extended thinking
# Add a tool and force its use
@@ -718,6 +737,7 @@ class AnthropicBaseLLMEntity(Entity):
# force tool use or disable text responses, so we add a hint to the
# system prompt instead. With extended thinking, the model should be
# smart enough to use the tool.
structure_name = slugify(structure_name)
model_args["tool_choice"] = ToolChoiceAutoParam(
type="auto",
)
@@ -725,22 +745,24 @@ class AnthropicBaseLLMEntity(Entity):
model_args["system"].append( # type: ignore[union-attr]
TextBlockParam(
type="text",
text=f"Claude MUST use the '{structure_name}' tool to provide the final answer instead of plain text.",
text=f"Claude MUST use the '{structure_name}' tool to provide "
"the final answer instead of plain text.",
)
)
tools.append(
ToolParam(
name=structure_name,
description="Use this tool to reply to the user",
input_schema=convert(
structure,
custom_serializer=chat_log.llm_api.custom_serializer
if chat_log.llm_api
else llm.selector_serializer,
),
if structure_name:
tools.append(
ToolParam(
name=structure_name,
description="Use this tool to reply to the user",
input_schema=convert(
structure,
custom_serializer=chat_log.llm_api.custom_serializer
if chat_log.llm_api
else llm.selector_serializer,
),
)
)
)
if tools:
model_args["tools"] = tools
@@ -761,7 +783,7 @@ class AnthropicBaseLLMEntity(Entity):
_transform_stream(
chat_log,
stream,
output_tool=structure_name if structure else None,
output_tool=structure_name or None,
),
)
]

16
tests/components/anthropic/conftest.py
View File

@@ -141,6 +141,22 @@ def mock_config_entry_with_web_search(
return mock_config_entry
@pytest.fixture
def mock_config_entry_with_no_structured_output(
hass: HomeAssistant, mock_config_entry: MockConfigEntry
) -> MockConfigEntry:
"""Mock a config entry with a model without structured outputs support."""
for subentry in mock_config_entry.subentries.values():
hass.config_entries.async_update_subentry(
mock_config_entry,
subentry,
data={
CONF_CHAT_MODEL: "claude-sonnet-4-0",
},
)
return mock_config_entry
@pytest.fixture
async def mock_init_component(
hass: HomeAssistant, mock_config_entry: MockConfigEntry

44
tests/components/anthropic/test_ai_task.py
View File

@@ -51,13 +51,13 @@ async def test_generate_data(
assert result.data == "The test data"
async def test_generate_structured_data(
async def test_generate_structured_data_legacy(
hass: HomeAssistant,
mock_config_entry: MockConfigEntry,
mock_config_entry_with_no_structured_output: MockConfigEntry,
mock_init_component,
mock_create_stream: AsyncMock,
) -> None:
"""Test AI Task structured data generation."""
"""Test AI Task structured data generation with legacy method."""
mock_create_stream.return_value = [
create_tool_use_block(
1,
@@ -88,13 +88,13 @@ async def test_generate_structured_data(
assert result.data == {"characters": ["Mario", "Luigi"]}
async def test_generate_invalid_structured_data(
async def test_generate_invalid_structured_data_legacy(
hass: HomeAssistant,
mock_config_entry: MockConfigEntry,
mock_config_entry_with_no_structured_output: MockConfigEntry,
mock_init_component,
mock_create_stream: AsyncMock,
) -> None:
"""Test AI Task with invalid JSON response."""
"""Test AI Task with invalid JSON response with legacy method."""
mock_create_stream.return_value = [
create_tool_use_block(
1,
@@ -126,6 +126,38 @@ async def test_generate_invalid_structured_data(
)
async def test_generate_structured_data(
hass: HomeAssistant,
mock_config_entry: MockConfigEntry,
mock_init_component,
mock_create_stream: AsyncMock,
) -> None:
"""Test AI Task structured data generation."""
mock_create_stream.return_value = [
create_content_block(0, ['{"charac', 'ters": ["Mario', '", "Luigi"]}'])
]
result = await ai_task.async_generate_data(
hass,
task_name="Test Task",
entity_id="ai_task.claude_ai_task",
instructions="Generate test data",
structure=vol.Schema(
{
vol.Required("characters"): selector.selector(
{
"text": {
"multiple": True,
}
}
)
},
),
)
assert result.data == {"characters": ["Mario", "Luigi"]}
async def test_generate_data_with_attachments(
hass: HomeAssistant,
mock_config_entry: MockConfigEntry,

10
tests/components/anthropic/test_config_flow.py
View File

@@ -296,7 +296,8 @@ async def test_subentry_web_search_user_location(
usage=types.Usage(input_tokens=100, output_tokens=100),
content=[
types.TextBlock(
type="text", text='"city": "San Francisco", "region": "California"}'
type="text",
text='{"city": "San Francisco", "region": "California"}',
)
],
),
@@ -313,12 +314,7 @@ async def test_subentry_web_search_user_location(
assert (
mock_create.call_args.kwargs["messages"][0]["content"] == "Where are the "
"following coordinates located: (37.7749, -122.4194)? Please respond only "
"with a JSON object using the following schema:\n"
"{'type': 'object', 'properties': {'city': {'type': 'string', 'description': "
"'Free text input for the city, e.g. `San Francisco`'}, 'region': {'type': "
"'string', 'description': 'Free text input for the region, e.g. `California`'"
"}}, 'required': []}"
"following coordinates located: (37.7749, -122.4194)?"
)
assert options["type"] is FlowResultType.ABORT
assert options["reason"] == "reconfigure_successful"