chore(cleanup)!: remove tool_runtime.rag_tool (#3871)

Kill the `builtin::rag` tool group completely since it is no longer
targeted. We use the Responses implementation for knowledge_search which
uses the `openai_vector_stores` pathway.

---------

Co-authored-by: github-actions[bot] <github-actions[bot]@users.noreply.github.com>

llama_stack/core/routers/tool_runtime.py

@@ -8,16 +8,8 @@ from typing import Any
 
 from llama_stack.apis.common.content_types import (
     URL,
-    InterleavedContent,
-)
-from llama_stack.apis.tools import (
-    ListToolDefsResponse,
-    RAGDocument,
-    RAGQueryConfig,
-    RAGQueryResult,
-    RAGToolRuntime,
-    ToolRuntime,
 )
+from llama_stack.apis.tools import ListToolDefsResponse, ToolRuntime
 from llama_stack.log import get_logger
 
 from ..routing_tables.toolgroups import ToolGroupsRoutingTable
@@ -26,36 +18,6 @@ logger = get_logger(name=__name__, category="core::routers")
 
 
 class ToolRuntimeRouter(ToolRuntime):
-    class RagToolImpl(RAGToolRuntime):
-        def __init__(
-            self,
-            routing_table: ToolGroupsRoutingTable,
-        ) -> None:
-            logger.debug("Initializing ToolRuntimeRouter.RagToolImpl")
-            self.routing_table = routing_table
-
-        async def query(
-            self,
-            content: InterleavedContent,
-            vector_store_ids: list[str],
-            query_config: RAGQueryConfig | None = None,
-        ) -> RAGQueryResult:
-            logger.debug(f"ToolRuntimeRouter.RagToolImpl.query: {vector_store_ids}")
-            provider = await self.routing_table.get_provider_impl("knowledge_search")
-            return await provider.query(content, vector_store_ids, query_config)
-
-        async def insert(
-            self,
-            documents: list[RAGDocument],
-            vector_store_id: str,
-            chunk_size_in_tokens: int = 512,
-        ) -> None:
-            logger.debug(
-                f"ToolRuntimeRouter.RagToolImpl.insert: {vector_store_id}, {len(documents)} documents, chunk_size={chunk_size_in_tokens}"
-            )
-            provider = await self.routing_table.get_provider_impl("insert_into_memory")
-            return await provider.insert(documents, vector_store_id, chunk_size_in_tokens)
-
     def __init__(
         self,
         routing_table: ToolGroupsRoutingTable,
@@ -63,11 +25,6 @@ class ToolRuntimeRouter(ToolRuntime):
         logger.debug("Initializing ToolRuntimeRouter")
         self.routing_table = routing_table
 
-        # HACK ALERT this should be in sync with "get_all_api_endpoints()"
-        self.rag_tool = self.RagToolImpl(routing_table)
-        for method in ("query", "insert"):
-            setattr(self, f"rag_tool.{method}", getattr(self.rag_tool, method))
-
     async def initialize(self) -> None:
         logger.debug("ToolRuntimeRouter.initialize")
         pass

llama_stack/core/server/routes.py

@@ -13,7 +13,6 @@ from aiohttp import hdrs
 from starlette.routing import Route
 
 from llama_stack.apis.datatypes import Api, ExternalApiSpec
-from llama_stack.apis.tools import RAGToolRuntime, SpecialToolGroup
 from llama_stack.core.resolver import api_protocol_map
 from llama_stack.schema_utils import WebMethod
 
@@ -25,33 +24,16 @@ RouteImpls = dict[str, PathImpl]
 RouteMatch = tuple[EndpointFunc, PathParams, str, WebMethod]
 
 
-def toolgroup_protocol_map():
-    return {
-        SpecialToolGroup.rag_tool: RAGToolRuntime,
-    }
-
-
 def get_all_api_routes(
     external_apis: dict[Api, ExternalApiSpec] | None = None,
 ) -> dict[Api, list[tuple[Route, WebMethod]]]:
     apis = {}
 
     protocols = api_protocol_map(external_apis)
-    toolgroup_protocols = toolgroup_protocol_map()
     for api, protocol in protocols.items():
         routes = []
         protocol_methods = inspect.getmembers(protocol, predicate=inspect.isfunction)
 
-        # HACK ALERT
-        if api == Api.tool_runtime:
-            for tool_group in SpecialToolGroup:
-                sub_protocol = toolgroup_protocols[tool_group]
-                sub_protocol_methods = inspect.getmembers(sub_protocol, predicate=inspect.isfunction)
-                for name, method in sub_protocol_methods:
-                    if not hasattr(method, "__webmethod__"):
-                        continue
-                    protocol_methods.append((f"{tool_group.value}.{name}", method))
-
         for name, method in protocol_methods:
             # Get all webmethods for this method (supports multiple decorators)
             webmethods = getattr(method, "__webmethods__", [])

llama_stack/core/stack.py

@@ -32,7 +32,7 @@ from llama_stack.apis.scoring_functions import ScoringFunctions
 from llama_stack.apis.shields import Shields
 from llama_stack.apis.synthetic_data_generation import SyntheticDataGeneration
 from llama_stack.apis.telemetry import Telemetry
-from llama_stack.apis.tools import RAGToolRuntime, ToolGroups, ToolRuntime
+from llama_stack.apis.tools import ToolGroups, ToolRuntime
 from llama_stack.apis.vector_io import VectorIO
 from llama_stack.core.conversations.conversations import ConversationServiceConfig, ConversationServiceImpl
 from llama_stack.core.datatypes import Provider, StackRunConfig, VectorStoresConfig
@@ -80,7 +80,6 @@ class LlamaStack(
     Inspect,
     ToolGroups,
     ToolRuntime,
-    RAGToolRuntime,
     Files,
     Prompts,
     Conversations,