feat(tools)!: substantial clean up of "Tool" related datatypes (#3627)

This is a sweeping change to clean up some gunk around our "Tool"
definitions.

First, we had two types `Tool` and `ToolDef`. The first of these was a
"Resource" type for the registry but we had stopped registering tools
inside the Registry long back (and only registered ToolGroups.) The
latter was for specifying tools for the Agents API. This PR removes the
former and adds an optional `toolgroup_id` field to the latter.

Secondly, as pointed out by @bbrowning in
https://github.com/llamastack/llama-stack/pull/3003#issuecomment-3245270132,
we were doing a lossy conversion from a full JSON schema from the MCP
tool specification into our ToolDefinition to send it to the model.
There is no necessity to do this -- we ourselves aren't doing any
execution at all but merely passing it to the chat completions API which
supports this. By doing this (and by doing it poorly), we encountered
limitations like not supporting array items, or not resolving $refs,
etc.

To fix this, we replaced the `parameters` field by `{ input_schema,
output_schema }` which can be full blown JSON schemas.

Finally, there were some types in our llama-related chat format
conversion which needed some cleanup. We are taking this opportunity to
clean those up.

This PR is a substantial breaking change to the API. However, given our
window for introducing breaking changes, this suits us just fine. I will
be landing a concurrent `llama-stack-client` change as well since API
shapes are changing.
This commit is contained in:
Ashwin Bharambe 2025-10-02 15:12:03 -07:00 committed by GitHub
parent 1f5003d50e
commit ef0736527d
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
179 changed files with 34186 additions and 9171 deletions

View file

@ -16,9 +16,8 @@ from llama_stack.apis.agents import (
)
from llama_stack.apis.common.responses import PaginatedResponse
from llama_stack.apis.inference import Inference
from llama_stack.apis.resource import ResourceType
from llama_stack.apis.safety import Safety
from llama_stack.apis.tools import ListToolsResponse, Tool, ToolGroups, ToolParameter, ToolRuntime
from llama_stack.apis.tools import ListToolDefsResponse, ToolDef, ToolGroups, ToolRuntime
from llama_stack.apis.vector_io import VectorIO
from llama_stack.providers.inline.agents.meta_reference.agent_instance import ChatAgent
from llama_stack.providers.inline.agents.meta_reference.agents import MetaReferenceAgentsImpl
@ -232,32 +231,26 @@ async def test_delete_agent(agents_impl, sample_agent_config):
async def test__initialize_tools(agents_impl, sample_agent_config):
# Mock tool_groups_api.list_tools()
agents_impl.tool_groups_api.list_tools.return_value = ListToolsResponse(
agents_impl.tool_groups_api.list_tools.return_value = ListToolDefsResponse(
data=[
Tool(
identifier="story_maker",
provider_id="model-context-protocol",
type=ResourceType.tool,
ToolDef(
name="story_maker",
toolgroup_id="mcp::my_mcp_server",
description="Make a story",
parameters=[
ToolParameter(
name="story_title",
parameter_type="string",
description="Title of the story",
required=True,
title="Story Title",
),
ToolParameter(
name="input_words",
parameter_type="array",
description="Input words",
required=False,
items={"type": "string"},
title="Input Words",
default=[],
),
],
input_schema={
"type": "object",
"properties": {
"story_title": {"type": "string", "description": "Title of the story", "title": "Story Title"},
"input_words": {
"type": "array",
"description": "Input words",
"items": {"type": "string"},
"title": "Input Words",
"default": [],
},
},
"required": ["story_title"],
},
)
]
)
@ -284,27 +277,27 @@ async def test__initialize_tools(agents_impl, sample_agent_config):
assert second_tool.tool_name == "story_maker"
assert second_tool.description == "Make a story"
parameters = second_tool.parameters
assert len(parameters) == 2
# Verify the input schema
input_schema = second_tool.input_schema
assert input_schema is not None
assert input_schema["type"] == "object"
properties = input_schema["properties"]
assert len(properties) == 2
# Verify a string property
story_title = parameters.get("story_title")
assert story_title is not None
assert story_title.param_type == "string"
assert story_title.description == "Title of the story"
assert story_title.required
assert story_title.items is None
assert story_title.title == "Story Title"
assert story_title.default is None
story_title = properties["story_title"]
assert story_title["type"] == "string"
assert story_title["description"] == "Title of the story"
assert story_title["title"] == "Story Title"
# Verify an array property
input_words = parameters.get("input_words")
assert input_words is not None
assert input_words.param_type == "array"
assert input_words.description == "Input words"
assert not input_words.required
assert input_words.items is not None
assert len(input_words.items) == 1
assert input_words.items.get("type") == "string"
assert input_words.title == "Input Words"
assert input_words.default == []
input_words = properties["input_words"]
assert input_words["type"] == "array"
assert input_words["description"] == "Input words"
assert input_words["items"]["type"] == "string"
assert input_words["title"] == "Input Words"
assert input_words["default"] == []
# Verify required fields
assert input_schema["required"] == ["story_title"]

View file

@ -39,7 +39,7 @@ from llama_stack.apis.inference import (
OpenAIResponseFormatJSONSchema,
OpenAIUserMessageParam,
)
from llama_stack.apis.tools.tools import Tool, ToolGroups, ToolInvocationResult, ToolParameter, ToolRuntime
from llama_stack.apis.tools.tools import ToolDef, ToolGroups, ToolInvocationResult, ToolRuntime
from llama_stack.core.access_control.access_control import default_policy
from llama_stack.core.datatypes import ResponsesStoreConfig
from llama_stack.providers.inline.agents.meta_reference.responses.openai_responses import (
@ -186,14 +186,15 @@ async def test_create_openai_response_with_string_input_with_tools(openai_respon
input_text = "What is the capital of Ireland?"
model = "meta-llama/Llama-3.1-8B-Instruct"
openai_responses_impl.tool_groups_api.get_tool.return_value = Tool(
identifier="web_search",
provider_id="client",
openai_responses_impl.tool_groups_api.get_tool.return_value = ToolDef(
name="web_search",
toolgroup_id="web_search",
description="Search the web for information",
parameters=[
ToolParameter(name="query", parameter_type="string", description="The query to search for", required=True)
],
input_schema={
"type": "object",
"properties": {"query": {"type": "string", "description": "The query to search for"}},
"required": ["query"],
},
)
openai_responses_impl.tool_runtime_api.invoke_tool.return_value = ToolInvocationResult(

View file

@ -138,8 +138,7 @@ async def test_tool_call_response(vllm_inference_adapter):
ToolCall(
call_id="foo",
tool_name="knowledge_search",
arguments={"query": "How many?"},
arguments_json='{"query": "How many?"}',
arguments='{"query": "How many?"}',
)
],
),
@ -263,7 +262,7 @@ async def test_tool_call_delta_streaming_arguments_dict():
assert chunks[1].event.event_type.value == "progress"
assert chunks[1].event.delta.type == "tool_call"
assert chunks[1].event.delta.parse_status.value == "succeeded"
assert chunks[1].event.delta.tool_call.arguments_json == '{"number": 28, "power": 3}'
assert chunks[1].event.delta.tool_call.arguments == '{"number": 28, "power": 3}'
assert chunks[2].event.event_type.value == "complete"
@ -339,11 +338,11 @@ async def test_multiple_tool_calls():
assert chunks[1].event.event_type.value == "progress"
assert chunks[1].event.delta.type == "tool_call"
assert chunks[1].event.delta.parse_status.value == "succeeded"
assert chunks[1].event.delta.tool_call.arguments_json == '{"number": 28, "power": 3}'
assert chunks[1].event.delta.tool_call.arguments == '{"number": 28, "power": 3}'
assert chunks[2].event.event_type.value == "progress"
assert chunks[2].event.delta.type == "tool_call"
assert chunks[2].event.delta.parse_status.value == "succeeded"
assert chunks[2].event.delta.tool_call.arguments_json == '{"first_number": 4, "second_number": 7}'
assert chunks[2].event.delta.tool_call.arguments == '{"first_number": 4, "second_number": 7}'
assert chunks[3].event.event_type.value == "complete"
@ -456,7 +455,7 @@ async def test_process_vllm_chat_completion_stream_response_tool_call_args_last_
assert chunks[-1].event.event_type == ChatCompletionResponseEventType.complete
assert chunks[-2].event.delta.type == "tool_call"
assert chunks[-2].event.delta.tool_call.tool_name == mock_tool_name
assert chunks[-2].event.delta.tool_call.arguments == mock_tool_arguments
assert chunks[-2].event.delta.tool_call.arguments == mock_tool_arguments_str
async def test_process_vllm_chat_completion_stream_response_no_finish_reason():
@ -468,7 +467,7 @@ async def test_process_vllm_chat_completion_stream_response_no_finish_reason():
mock_tool_name = "mock_tool"
mock_tool_arguments = {"arg1": 0, "arg2": 100}
mock_tool_arguments_str = '"{\\"arg1\\": 0, \\"arg2\\": 100}"'
mock_tool_arguments_str = json.dumps(mock_tool_arguments)
async def mock_stream():
mock_chunks = [
@ -508,7 +507,7 @@ async def test_process_vllm_chat_completion_stream_response_no_finish_reason():
assert chunks[-1].event.event_type == ChatCompletionResponseEventType.complete
assert chunks[-2].event.delta.type == "tool_call"
assert chunks[-2].event.delta.tool_call.tool_name == mock_tool_name
assert chunks[-2].event.delta.tool_call.arguments == mock_tool_arguments
assert chunks[-2].event.delta.tool_call.arguments == mock_tool_arguments_str
async def test_process_vllm_chat_completion_stream_response_tool_without_args():
@ -556,7 +555,7 @@ async def test_process_vllm_chat_completion_stream_response_tool_without_args():
assert chunks[-1].event.event_type == ChatCompletionResponseEventType.complete
assert chunks[-2].event.delta.type == "tool_call"
assert chunks[-2].event.delta.tool_call.tool_name == mock_tool_name
assert chunks[-2].event.delta.tool_call.arguments == {}
assert chunks[-2].event.delta.tool_call.arguments == "{}"
async def test_health_status_success(vllm_inference_adapter):

View file

@ -4,7 +4,7 @@
# This source code is licensed under the terms described in the LICENSE file in
# the root directory of this source tree.
from llama_stack.apis.tools import ToolDef, ToolParameter
from llama_stack.apis.tools import ToolDef
from llama_stack.providers.inline.agents.meta_reference.responses.streaming import (
convert_tooldef_to_chat_tool,
)
@ -20,15 +20,11 @@ def test_convert_tooldef_to_chat_tool_preserves_items_field():
tool_def = ToolDef(
name="test_tool",
description="A test tool with array parameter",
parameters=[
ToolParameter(
name="tags",
parameter_type="array",
description="List of tags",
required=True,
items={"type": "string"},
)
],
input_schema={
"type": "object",
"properties": {"tags": {"type": "array", "description": "List of tags", "items": {"type": "string"}}},
"required": ["tags"],
},
)
result = convert_tooldef_to_chat_tool(tool_def)

View file

@ -41,9 +41,7 @@ async def test_convert_message_to_openai_dict():
async def test_convert_message_to_openai_dict_with_tool_call():
message = CompletionMessage(
content="",
tool_calls=[
ToolCall(call_id="123", tool_name="test_tool", arguments_json='{"foo": "bar"}', arguments={"foo": "bar"})
],
tool_calls=[ToolCall(call_id="123", tool_name="test_tool", arguments='{"foo": "bar"}')],
stop_reason=StopReason.end_of_turn,
)
@ -65,8 +63,7 @@ async def test_convert_message_to_openai_dict_with_builtin_tool_call():
ToolCall(
call_id="123",
tool_name=BuiltinTool.brave_search,
arguments_json='{"foo": "bar"}',
arguments={"foo": "bar"},
arguments='{"foo": "bar"}',
)
],
stop_reason=StopReason.end_of_turn,
@ -202,8 +199,7 @@ async def test_convert_message_to_openai_dict_new_completion_message_with_tool_c
ToolCall(
call_id="call_123",
tool_name="get_weather",
arguments={"city": "Sligo"},
arguments_json='{"city": "Sligo"}',
arguments='{"city": "Sligo"}',
)
],
stop_reason=StopReason.end_of_turn,

View file

@ -0,0 +1,381 @@
# Copyright (c) Meta Platforms, Inc. and affiliates.
# All rights reserved.
#
# This source code is licensed under the terms described in the LICENSE file in
# the root directory of this source tree.
"""
Unit tests for OpenAI compatibility tool conversion.
Tests convert_tooldef_to_openai_tool with new JSON Schema approach.
"""
from llama_stack.models.llama.datatypes import BuiltinTool, ToolDefinition
from llama_stack.providers.utils.inference.openai_compat import convert_tooldef_to_openai_tool
class TestSimpleSchemaConversion:
"""Test basic schema conversions to OpenAI format."""
def test_simple_tool_conversion(self):
"""Test conversion of simple tool with basic input schema."""
tool = ToolDefinition(
tool_name="get_weather",
description="Get weather information",
input_schema={
"type": "object",
"properties": {"location": {"type": "string", "description": "City name"}},
"required": ["location"],
},
)
result = convert_tooldef_to_openai_tool(tool)
# Check OpenAI structure
assert result["type"] == "function"
assert "function" in result
function = result["function"]
assert function["name"] == "get_weather"
assert function["description"] == "Get weather information"
# Check parameters are passed through
assert "parameters" in function
assert function["parameters"] == tool.input_schema
assert function["parameters"]["type"] == "object"
assert "location" in function["parameters"]["properties"]
def test_tool_without_description(self):
"""Test tool conversion without description."""
tool = ToolDefinition(tool_name="test_tool", input_schema={"type": "object", "properties": {}})
result = convert_tooldef_to_openai_tool(tool)
assert result["function"]["name"] == "test_tool"
assert "description" not in result["function"]
assert "parameters" in result["function"]
def test_builtin_tool_conversion(self):
"""Test conversion of BuiltinTool enum."""
tool = ToolDefinition(
tool_name=BuiltinTool.code_interpreter,
description="Run Python code",
input_schema={"type": "object", "properties": {"code": {"type": "string"}}},
)
result = convert_tooldef_to_openai_tool(tool)
# BuiltinTool should be converted to its value
assert result["function"]["name"] == "code_interpreter"
class TestComplexSchemaConversion:
"""Test conversion of complex JSON Schema features."""
def test_schema_with_refs_and_defs(self):
"""Test that $ref and $defs are passed through to OpenAI."""
tool = ToolDefinition(
tool_name="book_flight",
description="Book a flight",
input_schema={
"type": "object",
"properties": {
"flight": {"$ref": "#/$defs/FlightInfo"},
"passengers": {"type": "array", "items": {"$ref": "#/$defs/Passenger"}},
"payment": {"$ref": "#/$defs/Payment"},
},
"required": ["flight", "passengers", "payment"],
"$defs": {
"FlightInfo": {
"type": "object",
"properties": {
"from": {"type": "string", "description": "Departure airport"},
"to": {"type": "string", "description": "Arrival airport"},
"date": {"type": "string", "format": "date"},
},
"required": ["from", "to", "date"],
},
"Passenger": {
"type": "object",
"properties": {"name": {"type": "string"}, "age": {"type": "integer", "minimum": 0}},
"required": ["name", "age"],
},
"Payment": {
"type": "object",
"properties": {
"method": {"type": "string", "enum": ["credit_card", "debit_card"]},
"amount": {"type": "number", "minimum": 0},
},
},
},
},
)
result = convert_tooldef_to_openai_tool(tool)
params = result["function"]["parameters"]
# Verify $defs are preserved
assert "$defs" in params
assert "FlightInfo" in params["$defs"]
assert "Passenger" in params["$defs"]
assert "Payment" in params["$defs"]
# Verify $ref are preserved
assert params["properties"]["flight"]["$ref"] == "#/$defs/FlightInfo"
assert params["properties"]["passengers"]["items"]["$ref"] == "#/$defs/Passenger"
assert params["properties"]["payment"]["$ref"] == "#/$defs/Payment"
# Verify nested schema details are preserved
assert params["$defs"]["FlightInfo"]["properties"]["date"]["format"] == "date"
assert params["$defs"]["Passenger"]["properties"]["age"]["minimum"] == 0
assert params["$defs"]["Payment"]["properties"]["method"]["enum"] == ["credit_card", "debit_card"]
def test_anyof_schema_conversion(self):
"""Test conversion of anyOf schemas."""
tool = ToolDefinition(
tool_name="flexible_input",
input_schema={
"type": "object",
"properties": {
"contact": {
"anyOf": [
{"type": "string", "format": "email"},
{"type": "string", "pattern": "^\\+?[0-9]{10,15}$"},
],
"description": "Email or phone number",
}
},
},
)
result = convert_tooldef_to_openai_tool(tool)
contact_schema = result["function"]["parameters"]["properties"]["contact"]
assert "anyOf" in contact_schema
assert len(contact_schema["anyOf"]) == 2
assert contact_schema["anyOf"][0]["format"] == "email"
assert "pattern" in contact_schema["anyOf"][1]
def test_nested_objects_conversion(self):
"""Test conversion of deeply nested objects."""
tool = ToolDefinition(
tool_name="nested_data",
input_schema={
"type": "object",
"properties": {
"user": {
"type": "object",
"properties": {
"profile": {
"type": "object",
"properties": {
"name": {"type": "string"},
"settings": {
"type": "object",
"properties": {"theme": {"type": "string", "enum": ["light", "dark"]}},
},
},
}
},
}
},
},
)
result = convert_tooldef_to_openai_tool(tool)
# Navigate deep structure
user_schema = result["function"]["parameters"]["properties"]["user"]
profile_schema = user_schema["properties"]["profile"]
settings_schema = profile_schema["properties"]["settings"]
theme_schema = settings_schema["properties"]["theme"]
assert theme_schema["enum"] == ["light", "dark"]
def test_array_schemas_with_constraints(self):
"""Test conversion of array schemas with constraints."""
tool = ToolDefinition(
tool_name="list_processor",
input_schema={
"type": "object",
"properties": {
"items": {
"type": "array",
"items": {
"type": "object",
"properties": {"id": {"type": "integer"}, "name": {"type": "string"}},
"required": ["id"],
},
"minItems": 1,
"maxItems": 100,
"uniqueItems": True,
}
},
},
)
result = convert_tooldef_to_openai_tool(tool)
items_schema = result["function"]["parameters"]["properties"]["items"]
assert items_schema["type"] == "array"
assert items_schema["minItems"] == 1
assert items_schema["maxItems"] == 100
assert items_schema["uniqueItems"] is True
assert items_schema["items"]["type"] == "object"
class TestOutputSchemaHandling:
"""Test that output_schema is correctly handled (or dropped) for OpenAI."""
def test_output_schema_is_dropped(self):
"""Test that output_schema is NOT included in OpenAI format (API limitation)."""
tool = ToolDefinition(
tool_name="calculator",
description="Perform calculation",
input_schema={"type": "object", "properties": {"x": {"type": "number"}, "y": {"type": "number"}}},
output_schema={"type": "object", "properties": {"result": {"type": "number"}}, "required": ["result"]},
)
result = convert_tooldef_to_openai_tool(tool)
# OpenAI doesn't support output schema
assert "outputSchema" not in result["function"]
assert "responseSchema" not in result["function"]
assert "output_schema" not in result["function"]
# But input schema should be present
assert "parameters" in result["function"]
assert result["function"]["parameters"] == tool.input_schema
def test_only_output_schema_no_input(self):
"""Test tool with only output_schema (unusual but valid)."""
tool = ToolDefinition(
tool_name="no_input_tool",
description="Tool with no inputs",
output_schema={"type": "object", "properties": {"timestamp": {"type": "string"}}},
)
result = convert_tooldef_to_openai_tool(tool)
# No parameters should be set if input_schema is None
# (or we might set an empty object schema - implementation detail)
assert "outputSchema" not in result["function"]
class TestEdgeCases:
"""Test edge cases and error conditions."""
def test_tool_with_no_schemas(self):
"""Test tool with neither input nor output schema."""
tool = ToolDefinition(tool_name="schemaless_tool", description="Tool without schemas")
result = convert_tooldef_to_openai_tool(tool)
assert result["function"]["name"] == "schemaless_tool"
assert result["function"]["description"] == "Tool without schemas"
# Implementation detail: might have no parameters or empty object
def test_empty_input_schema(self):
"""Test tool with empty object schema."""
tool = ToolDefinition(tool_name="no_params", input_schema={"type": "object", "properties": {}})
result = convert_tooldef_to_openai_tool(tool)
assert result["function"]["parameters"]["type"] == "object"
assert result["function"]["parameters"]["properties"] == {}
def test_schema_with_additional_properties(self):
"""Test that additionalProperties is preserved."""
tool = ToolDefinition(
tool_name="flexible_tool",
input_schema={
"type": "object",
"properties": {"known_field": {"type": "string"}},
"additionalProperties": True,
},
)
result = convert_tooldef_to_openai_tool(tool)
assert result["function"]["parameters"]["additionalProperties"] is True
def test_schema_with_pattern_properties(self):
"""Test that patternProperties is preserved."""
tool = ToolDefinition(
tool_name="pattern_tool",
input_schema={"type": "object", "patternProperties": {"^[a-z]+$": {"type": "string"}}},
)
result = convert_tooldef_to_openai_tool(tool)
assert "patternProperties" in result["function"]["parameters"]
def test_schema_identity(self):
"""Test that converted schema is identical to input (no lossy conversion)."""
original_schema = {
"type": "object",
"properties": {"complex": {"$ref": "#/$defs/Complex"}},
"$defs": {
"Complex": {
"type": "object",
"properties": {"nested": {"anyOf": [{"type": "string"}, {"type": "number"}]}},
}
},
"required": ["complex"],
"additionalProperties": False,
}
tool = ToolDefinition(tool_name="test", input_schema=original_schema)
result = convert_tooldef_to_openai_tool(tool)
# Converted parameters should be EXACTLY the same as input
assert result["function"]["parameters"] == original_schema
class TestConversionConsistency:
"""Test consistency across multiple conversions."""
def test_multiple_tools_with_shared_defs(self):
"""Test converting multiple tools that could share definitions."""
tool1 = ToolDefinition(
tool_name="tool1",
input_schema={
"type": "object",
"properties": {"data": {"$ref": "#/$defs/Data"}},
"$defs": {"Data": {"type": "object", "properties": {"x": {"type": "number"}}}},
},
)
tool2 = ToolDefinition(
tool_name="tool2",
input_schema={
"type": "object",
"properties": {"info": {"$ref": "#/$defs/Data"}},
"$defs": {"Data": {"type": "object", "properties": {"y": {"type": "string"}}}},
},
)
result1 = convert_tooldef_to_openai_tool(tool1)
result2 = convert_tooldef_to_openai_tool(tool2)
# Each tool maintains its own $defs independently
assert result1["function"]["parameters"]["$defs"]["Data"]["properties"]["x"]["type"] == "number"
assert result2["function"]["parameters"]["$defs"]["Data"]["properties"]["y"]["type"] == "string"
def test_conversion_is_pure(self):
"""Test that conversion doesn't modify the original tool."""
original_schema = {
"type": "object",
"properties": {"x": {"type": "string"}},
"$defs": {"T": {"type": "number"}},
}
tool = ToolDefinition(tool_name="test", input_schema=original_schema.copy())
# Convert
convert_tooldef_to_openai_tool(tool)
# Original tool should be unchanged
assert tool.input_schema == original_schema
assert "$defs" in tool.input_schema