feat: tool outputs metadata

Summary:

Allows tools to output metadata. This is useful for evaluating tool outputs, e.g. RAG tool will output document IDs, which can be used to score recall.

Will need to make a similar change on the client side to support ClientTool outputting metadata.

Test Plan:

LLAMA_STACK_CONFIG=fireworks pytest -s -v tests/client-sdk/agents/test_agents.py

docs/_static/llama-stack-spec.html

@@ -4454,6 +4454,31 @@
                     },
                     "content": {
                         "$ref": "#/components/schemas/InterleavedContent"
+                    },
+                    "metadata": {
+                        "type": "object",
+                        "additionalProperties": {
+                            "oneOf": [
+                                {
+                                    "type": "null"
+                                },
+                                {
+                                    "type": "boolean"
+                                },
+                                {
+                                    "type": "number"
+                                },
+                                {
+                                    "type": "string"
+                                },
+                                {
+                                    "type": "array"
+                                },
+                                {
+                                    "type": "object"
+                                }
+                            ]
+                        }
                     }
                 },
                 "additionalProperties": false,
@@ -6625,6 +6650,31 @@
                     },
                     "error_code": {
                         "type": "integer"
+                    },
+                    "metadata": {
+                        "type": "object",
+                        "additionalProperties": {
+                            "oneOf": [
+                                {
+                                    "type": "null"
+                                },
+                                {
+                                    "type": "boolean"
+                                },
+                                {
+                                    "type": "number"
+                                },
+                                {
+                                    "type": "string"
+                                },
+                                {
+                                    "type": "array"
+                                },
+                                {
+                                    "type": "object"
+                                }
+                            ]
+                        }
                     }
                 },
                 "additionalProperties": false,
@@ -7474,9 +7524,37 @@
                 "properties": {
                     "content": {
                         "$ref": "#/components/schemas/InterleavedContent"
+                    },
+                    "metadata": {
+                        "type": "object",
+                        "additionalProperties": {
+                            "oneOf": [
+                                {
+                                    "type": "null"
+                                },
+                                {
+                                    "type": "boolean"
+                                },
+                                {
+                                    "type": "number"
+                                },
+                                {
+                                    "type": "string"
+                                },
+                                {
+                                    "type": "array"
+                                },
+                                {
+                                    "type": "object"
+                                }
+                            ]
+                        }
                     }
                 },
                 "additionalProperties": false,
+                "required": [
+                    "metadata"
+                ],
                 "title": "RAGQueryResult"
             },
             "QueryChunksRequest": {

docs/_static/llama-stack-spec.yaml

@@ -2896,6 +2896,16 @@ components:
             - type: string
         content:
           $ref: '#/components/schemas/InterleavedContent'
+        metadata:
+          type: object
+          additionalProperties:
+            oneOf:
+              - type: 'null'
+              - type: boolean
+              - type: number
+              - type: string
+              - type: array
+              - type: object
       additionalProperties: false
       required:
         - call_id
@@ -4289,6 +4299,16 @@ components:
           type: string
         error_code:
           type: integer
+        metadata:
+          type: object
+          additionalProperties:
+            oneOf:
+              - type: 'null'
+              - type: boolean
+              - type: number
+              - type: string
+              - type: array
+              - type: object
       additionalProperties: false
       required:
         - content
@@ -4862,7 +4882,19 @@ components:
       properties:
         content:
           $ref: '#/components/schemas/InterleavedContent'
+        metadata:
+          type: object
+          additionalProperties:
+            oneOf:
+              - type: 'null'
+              - type: boolean
+              - type: number
+              - type: string
+              - type: array
+              - type: object
       additionalProperties: false
+      required:
+        - metadata
       title: RAGQueryResult
     QueryChunksRequest:
       type: object

llama_stack/apis/inference/inference.py

@@ -165,6 +165,7 @@ class ToolResponse(BaseModel):
     call_id: str
     tool_name: Union[BuiltinTool, str]
     content: InterleavedContent
+    metadata: Optional[Dict[str, Any]] = None
 
     @field_validator("tool_name", mode="before")
     @classmethod

llama_stack/apis/tools/rag_tool.py

@@ -26,6 +26,7 @@ class RAGDocument(BaseModel):
 @json_schema_type
 class RAGQueryResult(BaseModel):
     content: Optional[InterleavedContent] = None
+    metadata: Dict[str, Any] = Field(default_factory=dict)
 
 
 @json_schema_type

llama_stack/apis/tools/tools.py

@@ -72,6 +72,7 @@ class ToolInvocationResult(BaseModel):
     content: InterleavedContent
     error_message: Optional[str] = None
     error_code: Optional[int] = None
+    metadata: Optional[Dict[str, Any]] = None
 
 
 class ToolStore(Protocol):

llama_stack/providers/inline/agents/meta_reference/agent_instance.py

@@ -60,7 +60,7 @@
     UserMessage,
 )
 from llama_stack.apis.safety import Safety
-from llama_stack.apis.tools import RAGDocument, RAGQueryConfig, ToolGroups, ToolRuntime
+from llama_stack.apis.tools import RAGDocument, RAGQueryConfig, ToolGroups, ToolInvocationResult, ToolRuntime
 from llama_stack.apis.vector_io import VectorIO
 from llama_stack.models.llama.datatypes import BuiltinTool, ToolCall, ToolParamDefinition
 from llama_stack.providers.utils.kvstore import KVStore
@@ -456,6 +456,7 @@ async def _run(
                                         call_id="",
                                         tool_name=MEMORY_QUERY_TOOL,
                                         content=retrieved_context or [],
+                                        metadata=result.metadata,
                                     )
                                 ],
                             ),
@@ -650,13 +651,21 @@ async def _run(
                     },
                 ) as span:
                     tool_execution_start_time = datetime.now()
-                    result_messages = await execute_tool_call_maybe(
+                    tool_call = message.tool_calls[0]
+                    tool_result = await execute_tool_call_maybe(
                         self.tool_runtime_api,
                         session_id,
-                        [message],
+                        tool_call,
                         toolgroup_args,
                         tool_to_group,
                     )
+                    result_messages = [
+                        ToolResponseMessage(
+                            call_id=tool_call.call_id,
+                            tool_name=tool_call.tool_name,
+                            content=tool_result.content,
+                        )
+                    ]
                     assert len(result_messages) == 1, "Currently not supporting multiple messages"
                     result_message = result_messages[0]
                     span.set_attribute("output", result_message.model_dump_json())
@@ -675,6 +684,7 @@ async def _run(
                                         call_id=result_message.call_id,
                                         tool_name=result_message.tool_name,
                                         content=result_message.content,
+                                        metadata=tool_result.metadata,
                                     )
                                 ],
                                 started_at=tool_execution_start_time,
@@ -913,19 +923,10 @@ async def attachment_message(tempdir: str, urls: List[URL]) -> ToolResponseMessa
 async def execute_tool_call_maybe(
     tool_runtime_api: ToolRuntime,
     session_id: str,
-    messages: List[CompletionMessage],
+    tool_call: ToolCall,
     toolgroup_args: Dict[str, Dict[str, Any]],
     tool_to_group: Dict[str, str],
-) -> List[ToolResponseMessage]:
-    # While Tools.run interface takes a list of messages,
-    # All tools currently only run on a single message
-    # When this changes, we can drop this assert
-    # Whether to call tools on each message and aggregate
-    # or aggregate and call tool once, reamins to be seen.
-    assert len(messages) == 1, "Expected single message"
-    message = messages[0]
-
-    tool_call = message.tool_calls[0]
+) -> ToolInvocationResult:
     name = tool_call.tool_name
     group_name = tool_to_group.get(name, None)
     if group_name is None:
@@ -946,14 +947,7 @@ async def execute_tool_call_maybe(
             **tool_call_args,
         ),
     )
-
-    return [
-        ToolResponseMessage(
-            call_id=tool_call.call_id,
-            tool_name=tool_call.tool_name,
-            content=result.content,
-        )
-    ]
+    return result
 
 
 def _interpret_content_as_attachment(

llama_stack/providers/inline/tool_runtime/rag/memory.py

@@ -119,10 +119,10 @@ async def query(
 
         # sort by score
         chunks, scores = zip(*sorted(zip(chunks, scores, strict=False), key=lambda x: x[1], reverse=True), strict=False)
-
+        chunks = chunks[: query_config.max_chunks]
         tokens = 0
         picked = []
-        for c in chunks[: query_config.max_chunks]:
+        for c in chunks:
             metadata = c.metadata
             tokens += metadata["token_count"]
             if tokens > query_config.max_tokens_in_context:
@@ -146,6 +146,9 @@ async def query(
                     text="\n=== END-RETRIEVED-CONTEXT ===\n",
                 ),
             ],
+            metadata={
+                "document_ids": [c.metadata["document_id"] for c in chunks[: len(picked)]],
+            },
         )
 
     async def list_runtime_tools(

tests/client-sdk/agents/test_agents.py

@@ -453,6 +453,7 @@ def test_rag_agent(llama_stack_client, agent_config):
         vector_db_id=vector_db_id,
         embedding_model="all-MiniLM-L6-v2",
         embedding_dimension=384,
+        provider_id="faiss",
     )
     llama_stack_client.tool_runtime.rag_tool.insert(
         documents=documents,
@@ -488,11 +489,13 @@ def test_rag_agent(llama_stack_client, agent_config):
         response = rag_agent.create_turn(
             messages=[{"role": "user", "content": prompt}],
             session_id=session_id,
+            stream=False,
         )
-        logs = [str(log) for log in EventLogger().log(response) if log is not None]
-        logs_str = "".join(logs)
-        assert "Tool:query_from_memory" in logs_str
-        assert expected_kw in logs_str.lower()
+        # rag is called
+        assert response.steps[0].tool_calls[0].tool_name == "query_from_memory"
+        # document ids are present in metadata
+        assert "num-0" in response.steps[0].tool_responses[0].metadata["document_ids"]
+        assert expected_kw in response.output_message.content
 
 
 def test_rag_and_code_agent(llama_stack_client, agent_config):