grasp_agents 0.5.10__py3-none-any.whl → 0.5.11__py3-none-any.whl

grasp_agents/cloud_llm.py

@@ -61,7 +61,6 @@ LLMRateLimiter = RateLimiterC[
 
 @dataclass(frozen=True)
 class CloudLLM(LLM[SettingsT_co, ConvertT_co], Generic[SettingsT_co, ConvertT_co]):
-    # Make this field keyword-only to avoid ordering issues with inherited defaulted fields
     api_provider: APIProvider | None = None
     llm_settings: SettingsT_co | None = None
     rate_limiter: LLMRateLimiter | None = None

grasp_agents/llm_agent.py

@@ -42,6 +42,7 @@ class OutputParser(Protocol[_InT_contra, _OutT_co, CtxT]):
         *,
         in_args: _InT_contra | None,
         ctx: RunContext[CtxT],
+        call_id: str,
     ) -> _OutT_co: ...
 
 
@@ -169,10 +170,15 @@ class LLMAgent(
         in_args: InT | None = None,
         sys_prompt: LLMPrompt | None = None,
         ctx: RunContext[Any],
+        call_id: str,
     ) -> None:
         if self.memory_preparator:
             return self.memory_preparator(
-                memory=memory, in_args=in_args, sys_prompt=sys_prompt, ctx=ctx
+                memory=memory,
+                in_args=in_args,
+                sys_prompt=sys_prompt,
+                ctx=ctx,
+                call_id=call_id,
             )
 
     def _memorize_inputs(
@@ -182,8 +188,11 @@ class LLMAgent(
         chat_inputs: LLMPrompt | Sequence[str | ImageData] | None = None,
         in_args: InT | None = None,
         ctx: RunContext[CtxT],
+        call_id: str,
     ) -> tuple[SystemMessage | None, UserMessage | None]:
-        formatted_sys_prompt = self._prompt_builder.build_system_prompt(ctx=ctx)
+        formatted_sys_prompt = self._prompt_builder.build_system_prompt(
+            ctx=ctx, call_id=call_id
+        )
 
         system_message: SystemMessage | None = None
         if self._reset_memory_on_run or memory.is_empty:
@@ -192,24 +201,22 @@ class LLMAgent(
                 system_message = cast("SystemMessage", memory.message_history[0])
         else:
             self._prepare_memory(
-                memory=memory, in_args=in_args, sys_prompt=formatted_sys_prompt, ctx=ctx
+                memory=memory,
+                in_args=in_args,
+                sys_prompt=formatted_sys_prompt,
+                ctx=ctx,
+                call_id=call_id,
             )
 
         input_message = self._prompt_builder.build_input_message(
-            chat_inputs=chat_inputs, in_args=in_args, ctx=ctx
+            chat_inputs=chat_inputs, in_args=in_args, ctx=ctx, call_id=call_id
         )
         if input_message:
             memory.update([input_message])
 
         return system_message, input_message
 
-    def _parse_output_default(
-        self,
-        conversation: Messages,
-        *,
-        in_args: InT | None = None,
-        ctx: RunContext[CtxT],
-    ) -> OutT:
+    def parse_output_default(self, conversation: Messages) -> OutT:
         return validate_obj_from_json_or_py_string(
             str(conversation[-1].content or ""),
             schema=self._out_type,
@@ -223,15 +230,14 @@ class LLMAgent(
         *,
         in_args: InT | None = None,
         ctx: RunContext[CtxT],
+        call_id: str,
     ) -> OutT:
         if self.output_parser:
             return self.output_parser(
-                conversation=conversation, in_args=in_args, ctx=ctx
+                conversation=conversation, in_args=in_args, ctx=ctx, call_id=call_id
             )
 
-        return self._parse_output_default(
-            conversation=conversation, in_args=in_args, ctx=ctx
-        )
+        return self.parse_output_default(conversation)
 
     async def _process(
         self,
@@ -239,24 +245,28 @@ class LLMAgent(
         *,
         in_args: InT | None = None,
         memory: LLMAgentMemory,
-        call_id: str,
         ctx: RunContext[CtxT],
+        call_id: str,
     ) -> OutT:
         system_message, input_message = self._memorize_inputs(
             memory=memory,
             chat_inputs=chat_inputs,
             in_args=in_args,
             ctx=ctx,
+            call_id=call_id,
         )
         if system_message:
-            self._print_messages([system_message], call_id=call_id, ctx=ctx)
+            self._print_messages([system_message], ctx=ctx, call_id=call_id)
         if input_message:
-            self._print_messages([input_message], call_id=call_id, ctx=ctx)
+            self._print_messages([input_message], ctx=ctx, call_id=call_id)
 
-        await self._policy_executor.execute(memory, call_id=call_id, ctx=ctx)
+        await self._policy_executor.execute(memory, ctx=ctx, call_id=call_id)
 
         return self._parse_output(
-            conversation=memory.message_history, in_args=in_args, ctx=ctx
+            conversation=memory.message_history,
+            in_args=in_args,
+            ctx=ctx,
+            call_id=call_id,
         )
 
     async def _process_stream(
@@ -265,41 +275,45 @@ class LLMAgent(
         *,
         in_args: InT | None = None,
         memory: LLMAgentMemory,
-        call_id: str,
         ctx: RunContext[CtxT],
+        call_id: str,
     ) -> AsyncIterator[Event[Any]]:
         system_message, input_message = self._memorize_inputs(
             memory=memory,
             chat_inputs=chat_inputs,
             in_args=in_args,
             ctx=ctx,
+            call_id=call_id,
         )
         if system_message:
-            self._print_messages([system_message], call_id=call_id, ctx=ctx)
+            self._print_messages([system_message], ctx=ctx, call_id=call_id)
             yield SystemMessageEvent(
                 data=system_message, proc_name=self.name, call_id=call_id
             )
         if input_message:
-            self._print_messages([input_message], call_id=call_id, ctx=ctx)
+            self._print_messages([input_message], ctx=ctx, call_id=call_id)
             yield UserMessageEvent(
                 data=input_message, proc_name=self.name, call_id=call_id
             )
 
         async for event in self._policy_executor.execute_stream(
-            memory, call_id=call_id, ctx=ctx
+            memory, ctx=ctx, call_id=call_id
         ):
             yield event
 
         output = self._parse_output(
-            conversation=memory.message_history, in_args=in_args, ctx=ctx
+            conversation=memory.message_history,
+            in_args=in_args,
+            ctx=ctx,
+            call_id=call_id,
         )
         yield ProcPayloadOutputEvent(data=output, proc_name=self.name, call_id=call_id)
 
     def _print_messages(
         self,
         messages: Sequence[Message],
-        call_id: str,
         ctx: RunContext[CtxT],
+        call_id: str,
     ) -> None:
         if ctx and ctx.printer:
             ctx.printer.print_messages(messages, agent_name=self.name, call_id=call_id)
@@ -328,31 +342,45 @@ class LLMAgent(
         if cur_cls.memory_manager is not base_cls.memory_manager:
             self._policy_executor.memory_manager = self.memory_manager
 
-    def system_prompt_builder(self, ctx: RunContext[CtxT]) -> str | None:
+    def system_prompt_builder(self, ctx: RunContext[CtxT], call_id: str) -> str | None:
         if self._prompt_builder.system_prompt_builder is not None:
-            return self._prompt_builder.system_prompt_builder(ctx=ctx)
+            return self._prompt_builder.system_prompt_builder(ctx=ctx, call_id=call_id)
         raise NotImplementedError("System prompt builder is not implemented.")
 
-    def input_content_builder(self, in_args: InT, ctx: RunContext[CtxT]) -> Content:
+    def input_content_builder(
+        self, in_args: InT, ctx: RunContext[CtxT], call_id: str
+    ) -> Content:
         if self._prompt_builder.input_content_builder is not None:
-            return self._prompt_builder.input_content_builder(in_args=in_args, ctx=ctx)
+            return self._prompt_builder.input_content_builder(
+                in_args=in_args, ctx=ctx, call_id=call_id
+            )
         raise NotImplementedError("Input content builder is not implemented.")
 
     def tool_call_loop_terminator(
-        self, conversation: Messages, *, ctx: RunContext[CtxT], **kwargs: Any
+        self,
+        conversation: Messages,
+        *,
+        ctx: RunContext[CtxT],
+        call_id: str,
+        **kwargs: Any,
     ) -> bool:
         if self._policy_executor.tool_call_loop_terminator is not None:
             return self._policy_executor.tool_call_loop_terminator(
-                conversation=conversation, ctx=ctx, **kwargs
+                conversation=conversation, ctx=ctx, call_id=call_id, **kwargs
             )
         raise NotImplementedError("Tool call loop terminator is not implemented.")
 
     def memory_manager(
-        self, memory: LLMAgentMemory, *, ctx: RunContext[CtxT], **kwargs: Any
+        self,
+        memory: LLMAgentMemory,
+        *,
+        ctx: RunContext[CtxT],
+        call_id: str,
+        **kwargs: Any,
     ) -> None:
         if self._policy_executor.memory_manager is not None:
             return self._policy_executor.memory_manager(
-                memory=memory, ctx=ctx, **kwargs
+                memory=memory, ctx=ctx, call_id=call_id, **kwargs
             )
         raise NotImplementedError("Memory manager is not implemented.")
 

grasp_agents/llm_agent_memory.py

@@ -17,6 +17,7 @@ class MemoryPreparator(Protocol):
         in_args: Any | None,
         sys_prompt: LLMPrompt | None,
         ctx: RunContext[Any],
+        call_id: str,
     ) -> None: ...
 
 

grasp_agents/llm_policy_executor.py

@@ -37,6 +37,7 @@ class ToolCallLoopTerminator(Protocol[CtxT]):
         conversation: Messages,
         *,
         ctx: RunContext[CtxT],
+        call_id: str,
         **kwargs: Any,
     ) -> bool: ...
 
@@ -47,6 +48,7 @@ class MemoryManager(Protocol[CtxT]):
         memory: LLMAgentMemory,
         *,
         ctx: RunContext[CtxT],
+        call_id: str,
         **kwargs: Any,
     ) -> None: ...
 
@@ -123,10 +125,13 @@ class LLMPolicyExecutor(Generic[CtxT]):
         conversation: Messages,
         *,
         ctx: RunContext[CtxT],
+        call_id: str,
         **kwargs: Any,
     ) -> bool:
         if self.tool_call_loop_terminator:
-            return self.tool_call_loop_terminator(conversation, ctx=ctx, **kwargs)
+            return self.tool_call_loop_terminator(
+                conversation, ctx=ctx, call_id=call_id, **kwargs
+            )
 
         return False
 
@@ -136,18 +141,19 @@ class LLMPolicyExecutor(Generic[CtxT]):
         memory: LLMAgentMemory,
         *,
         ctx: RunContext[CtxT],
+        call_id: str,
         **kwargs: Any,
     ) -> None:
         if self.memory_manager:
-            self.memory_manager(memory=memory, ctx=ctx, **kwargs)
+            self.memory_manager(memory=memory, ctx=ctx, call_id=call_id, **kwargs)
 
     async def generate_message(
         self,
         memory: LLMAgentMemory,
         *,
-        call_id: str,
         tool_choice: ToolChoice | None = None,
         ctx: RunContext[CtxT],
+        call_id: str,
     ) -> AssistantMessage:
         completion = await self.llm.generate_completion(
             memory.message_history,
@@ -161,7 +167,7 @@ class LLMPolicyExecutor(Generic[CtxT]):
         )
         memory.update(completion.messages)
         self._process_completion(
-            completion, call_id=call_id, ctx=ctx, print_messages=True
+            completion, ctx=ctx, call_id=call_id, print_messages=True
         )
 
         return completion.messages[0]
@@ -170,9 +176,9 @@ class LLMPolicyExecutor(Generic[CtxT]):
         self,
         memory: LLMAgentMemory,
         *,
-        call_id: str,
         tool_choice: ToolChoice | None = None,
         ctx: RunContext[CtxT],
+        call_id: str,
     ) -> AsyncIterator[
         CompletionChunkEvent[CompletionChunk]
         | CompletionEvent
@@ -207,22 +213,22 @@ class LLMPolicyExecutor(Generic[CtxT]):
         memory.update(completion.messages)
 
         self._process_completion(
-            completion, call_id=call_id, print_messages=True, ctx=ctx
+            completion, print_messages=True, ctx=ctx, call_id=call_id
         )
 
     async def call_tools(
         self,
         calls: Sequence[ToolCall],
         memory: LLMAgentMemory,
-        call_id: str,
         ctx: RunContext[CtxT],
+        call_id: str,
     ) -> Sequence[ToolMessage]:
         # TODO: Add image support
         corouts: list[Coroutine[Any, Any, BaseModel]] = []
         for call in calls:
             tool = self.tools[call.tool_name]
             args = json.loads(call.tool_arguments)
-            corouts.append(tool(call_id=call_id, ctx=ctx, **args))
+            corouts.append(tool(ctx=ctx, call_id=call_id, **args))
 
         outs = await asyncio.gather(*corouts)
         tool_messages = list(
@@ -242,11 +248,11 @@ class LLMPolicyExecutor(Generic[CtxT]):
         self,
         calls: Sequence[ToolCall],
         memory: LLMAgentMemory,
-        call_id: str,
         ctx: RunContext[CtxT],
+        call_id: str,
     ) -> AsyncIterator[ToolMessageEvent]:
         tool_messages = await self.call_tools(
-            calls, memory=memory, call_id=call_id, ctx=ctx
+            calls, memory=memory, ctx=ctx, call_id=call_id
         )
         for tool_message, call in zip(tool_messages, calls, strict=True):
             yield ToolMessageEvent(
@@ -271,7 +277,7 @@ class LLMPolicyExecutor(Generic[CtxT]):
                 return final_answer_message
 
     async def _generate_final_answer(
-        self, memory: LLMAgentMemory, call_id: str, ctx: RunContext[CtxT]
+        self, memory: LLMAgentMemory, ctx: RunContext[CtxT], call_id: str
     ) -> AssistantMessage:
         user_message = UserMessage.from_text(
             "Exceeded the maximum number of turns: provide a final answer now!"
@@ -284,7 +290,7 @@ class LLMPolicyExecutor(Generic[CtxT]):
 
         tool_choice = NamedToolChoice(name=self._final_answer_tool.name)
         await self.generate_message(
-            memory, tool_choice=tool_choice, call_id=call_id, ctx=ctx
+            memory, tool_choice=tool_choice, ctx=ctx, call_id=call_id
         )
 
         final_answer_message = self._extract_final_answer_from_tool_calls(memory=memory)
@@ -294,7 +300,7 @@ class LLMPolicyExecutor(Generic[CtxT]):
         return final_answer_message
 
     async def _generate_final_answer_stream(
-        self, memory: LLMAgentMemory, call_id: str, ctx: RunContext[CtxT]
+        self, memory: LLMAgentMemory, ctx: RunContext[CtxT], call_id: str
     ) -> AsyncIterator[Event[Any]]:
         user_message = UserMessage.from_text(
             "Exceeded the maximum number of turns: provide a final answer now!",
@@ -310,7 +316,7 @@ class LLMPolicyExecutor(Generic[CtxT]):
 
         tool_choice = NamedToolChoice(name=self._final_answer_tool.name)
         async for event in self.generate_message_stream(
-            memory, tool_choice=tool_choice, call_id=call_id, ctx=ctx
+            memory, tool_choice=tool_choice, ctx=ctx, call_id=call_id
         ):
             yield event
 
@@ -322,7 +328,7 @@ class LLMPolicyExecutor(Generic[CtxT]):
         )
 
     async def execute(
-        self, memory: LLMAgentMemory, call_id: str, ctx: RunContext[CtxT]
+        self, memory: LLMAgentMemory, ctx: RunContext[CtxT], call_id: str
     ) -> AssistantMessage | Sequence[AssistantMessage]:
         # 1. Generate the first message:
         #    In ReAct mode, we generate the first message without tool calls
@@ -332,7 +338,7 @@ class LLMPolicyExecutor(Generic[CtxT]):
         if self.tools:
             tool_choice = "none" if self._react_mode else "auto"
         gen_message = await self.generate_message(
-            memory, tool_choice=tool_choice, call_id=call_id, ctx=ctx
+            memory, tool_choice=tool_choice, ctx=ctx, call_id=call_id
         )
         if not self.tools:
             return gen_message
@@ -345,7 +351,7 @@ class LLMPolicyExecutor(Generic[CtxT]):
             # If a final answer is not provided via a tool call, we use
             # _terminate_tool_call_loop to determine whether to exit the loop.
             if not self._final_answer_as_tool_call and self._terminate_tool_call_loop(
-                memory.message_history, ctx=ctx, num_turns=turns
+                memory.message_history, ctx=ctx, call_id=call_id, num_turns=turns
             ):
                 return gen_message
 
@@ -364,7 +370,7 @@ class LLMPolicyExecutor(Generic[CtxT]):
                 # Otherwise, we simply return the last generated message.
                 if self._final_answer_as_tool_call:
                     final_answer = await self._generate_final_answer(
-                        memory, call_id=call_id, ctx=ctx
+                        memory, ctx=ctx, call_id=call_id
                     )
                 else:
                     final_answer = gen_message
@@ -377,11 +383,11 @@ class LLMPolicyExecutor(Generic[CtxT]):
 
             if gen_message.tool_calls:
                 await self.call_tools(
-                    gen_message.tool_calls, memory=memory, call_id=call_id, ctx=ctx
+                    gen_message.tool_calls, memory=memory, ctx=ctx, call_id=call_id
                 )
 
             # Apply memory management (e.g. compacting or pruning memory)
-            self._manage_memory(memory, ctx=ctx, num_turns=turns)
+            self._manage_memory(memory, ctx=ctx, call_id=call_id, num_turns=turns)
 
             # 4. Generate the next message based on the updated memory.
             #    In ReAct mode, we set tool_choice to "none" if we just called tools,
@@ -396,7 +402,7 @@ class LLMPolicyExecutor(Generic[CtxT]):
                 tool_choice = "required"
 
             gen_message = await self.generate_message(
-                memory, tool_choice=tool_choice, call_id=call_id, ctx=ctx
+                memory, tool_choice=tool_choice, ctx=ctx, call_id=call_id
             )
 
             turns += 1
@@ -404,13 +410,13 @@ class LLMPolicyExecutor(Generic[CtxT]):
     async def execute_stream(
         self,
         memory: LLMAgentMemory,
-        call_id: str,
         ctx: RunContext[CtxT],
+        call_id: str,
     ) -> AsyncIterator[Event[Any]]:
         tool_choice: ToolChoice = "none" if self._react_mode else "auto"
         gen_message: AssistantMessage | None = None
         async for event in self.generate_message_stream(
-            memory, tool_choice=tool_choice, call_id=call_id, ctx=ctx
+            memory, tool_choice=tool_choice, ctx=ctx, call_id=call_id
         ):
             if isinstance(event, GenMessageEvent):
                 gen_message = event.data
@@ -425,7 +431,7 @@ class LLMPolicyExecutor(Generic[CtxT]):
 
         while True:
             if not self._final_answer_as_tool_call and self._terminate_tool_call_loop(
-                memory.message_history, ctx=ctx, num_turns=turns
+                memory.message_history, ctx=ctx, call_id=call_id, num_turns=turns
             ):
                 return
 
@@ -444,7 +450,7 @@ class LLMPolicyExecutor(Generic[CtxT]):
             if turns >= self.max_turns:
                 if self._final_answer_as_tool_call:
                     async for event in self._generate_final_answer_stream(
-                        memory, call_id=call_id, ctx=ctx
+                        memory, ctx=ctx, call_id=call_id
                     ):
                         yield event
                 logger.info(
@@ -459,11 +465,11 @@ class LLMPolicyExecutor(Generic[CtxT]):
                     )
 
                 async for event in self.call_tools_stream(
-                    gen_message.tool_calls, memory=memory, call_id=call_id, ctx=ctx
+                    gen_message.tool_calls, memory=memory, ctx=ctx, call_id=call_id
                 ):
                     yield event
 
-            self._manage_memory(memory, ctx=ctx, num_turns=turns)
+            self._manage_memory(memory, ctx=ctx, call_id=call_id, num_turns=turns)
 
             if self._react_mode and gen_message.tool_calls:
                 tool_choice = "none"
@@ -473,7 +479,7 @@ class LLMPolicyExecutor(Generic[CtxT]):
                 tool_choice = "required"
 
             async for event in self.generate_message_stream(
-                memory, tool_choice=tool_choice, call_id=call_id, ctx=ctx
+                memory, tool_choice=tool_choice, ctx=ctx, call_id=call_id
             ):
                 yield event
                 if isinstance(event, GenMessageEvent):
@@ -493,8 +499,8 @@ class LLMPolicyExecutor(Generic[CtxT]):
                 self,
                 inp: BaseModel,
                 *,
-                call_id: str | None = None,
                 ctx: RunContext[Any] | None = None,
+                call_id: str | None = None,
             ) -> None:
                 return None
 
@@ -504,9 +510,9 @@ class LLMPolicyExecutor(Generic[CtxT]):
         self,
         completion: Completion,
         *,
-        call_id: str,
         print_messages: bool = False,
         ctx: RunContext[CtxT],
+        call_id: str,
     ) -> None:
         ctx.completions[self.agent_name].append(completion)
         ctx.usage_tracker.update(

grasp_agents/openai/openai_llm.py

@@ -60,7 +60,7 @@ def get_openai_compatible_providers() -> list[APIProvider]:
 
 
 class OpenAILLMSettings(CloudLLMSettings, total=False):
-    reasoning_effort: Literal["low", "medium", "high"] | None
+    reasoning_effort: Literal["disable", "minimal", "low", "medium", "high"] | None
 
     parallel_tool_calls: bool
 

grasp_agents/processors/base_processor.py

@@ -106,7 +106,7 @@ _OutT_contra = TypeVar("_OutT_contra", contravariant=True)
 
 class RecipientSelector(Protocol[_OutT_contra, CtxT]):
     def __call__(
-        self, output: _OutT_contra, ctx: RunContext[CtxT]
+        self, output: _OutT_contra, *, ctx: RunContext[CtxT]
     ) -> Sequence[ProcName] | None: ...
 
 

grasp_agents/processors/parallel_processor.py

@@ -114,7 +114,7 @@ class ParallelProcessor(
         ctx: RunContext[CtxT] | None = None,
     ) -> Packet[OutT]:
         call_id = self._generate_call_id(call_id)
-        ctx = RunContext[CtxT](state=None) if ctx is None else ctx  # type: ignore
+        ctx = ctx or RunContext[CtxT](state=None)  # type: ignore
 
         val_in_args = self._validate_inputs(
             call_id=call_id,
@@ -223,7 +223,7 @@ class ParallelProcessor(
         ctx: RunContext[CtxT] | None = None,
     ) -> AsyncIterator[Event[Any]]:
         call_id = self._generate_call_id(call_id)
-        ctx = RunContext[CtxT](state=None) if ctx is None else ctx  # type: ignore
+        ctx = ctx or RunContext[CtxT](state=None)  # type: ignore
 
         val_in_args = self._validate_inputs(
             call_id=call_id,

grasp_agents/processors/processor.py

@@ -105,7 +105,7 @@ class Processor(BaseProcessor[InT, OutT, MemT, CtxT], Generic[InT, OutT, MemT, C
         call_id: str | None = None,
         ctx: RunContext[CtxT] | None = None,
     ) -> Packet[OutT]:
-        ctx = RunContext[CtxT](state=None) if ctx is None else ctx  # type: ignore
+        ctx = ctx or RunContext[CtxT](state=None)  # type: ignore
 
         val_in_args, memory, call_id = self._preprocess(
             chat_inputs=chat_inputs,
@@ -136,7 +136,7 @@ class Processor(BaseProcessor[InT, OutT, MemT, CtxT], Generic[InT, OutT, MemT, C
         call_id: str | None = None,
         ctx: RunContext[CtxT] | None = None,
     ) -> AsyncIterator[Event[Any]]:
-        ctx = RunContext[CtxT](state=None) if ctx is None else ctx  # type: ignore
+        ctx = ctx or RunContext[CtxT](state=None)  # type: ignore
 
         val_in_args, memory, call_id = self._preprocess(
             chat_inputs=chat_inputs,

grasp_agents/prompt_builder.py

@@ -15,11 +15,22 @@ _InT_contra = TypeVar("_InT_contra", contravariant=True)
 
 
 class SystemPromptBuilder(Protocol[CtxT]):
-    def __call__(self, ctx: RunContext[CtxT]) -> str | None: ...
+    def __call__(
+        self,
+        *,
+        ctx: RunContext[CtxT],
+        call_id: str,
+    ) -> str | None: ...
 
 
 class InputContentBuilder(Protocol[_InT_contra, CtxT]):
-    def __call__(self, in_args: _InT_contra, ctx: RunContext[CtxT]) -> Content: ...
+    def __call__(
+        self,
+        in_args: _InT_contra,
+        *,
+        ctx: RunContext[CtxT],
+        call_id: str,
+    ) -> Content: ...
 
 
 PromptArgumentType: TypeAlias = str | bool | int | ImageData
@@ -43,9 +54,9 @@ class PromptBuilder(AutoInstanceAttributesMixin, Generic[InT, CtxT]):
         self._in_args_type_adapter: TypeAdapter[InT] = TypeAdapter(self._in_type)
 
     @final
-    def build_system_prompt(self, ctx: RunContext[CtxT]) -> str | None:
+    def build_system_prompt(self, ctx: RunContext[CtxT], call_id: str) -> str | None:
         if self.system_prompt_builder:
-            return self.system_prompt_builder(ctx=ctx)
+            return self.system_prompt_builder(ctx=ctx, call_id=call_id)
 
         return self.sys_prompt
 
@@ -71,7 +82,7 @@ class PromptBuilder(AutoInstanceAttributesMixin, Generic[InT, CtxT]):
 
     @final
     def _build_input_content(
-        self, in_args: InT | None, ctx: RunContext[CtxT]
+        self, in_args: InT | None, ctx: RunContext[CtxT], call_id: str
     ) -> Content:
         if in_args is None and self._in_type is not type(None):
             raise InputPromptBuilderError(
@@ -83,7 +94,9 @@ class PromptBuilder(AutoInstanceAttributesMixin, Generic[InT, CtxT]):
 
         val_in_args = self._validate_input_args(in_args)
         if self.input_content_builder:
-            return self.input_content_builder(in_args=val_in_args, ctx=ctx)
+            return self.input_content_builder(
+                in_args=val_in_args, ctx=ctx, call_id=call_id
+            )
 
         if issubclass(self._in_type, BaseModel) and isinstance(val_in_args, BaseModel):
             val_in_args_map = self._format_pydantic_prompt_args(val_in_args)
@@ -102,6 +115,7 @@ class PromptBuilder(AutoInstanceAttributesMixin, Generic[InT, CtxT]):
         chat_inputs: LLMPrompt | Sequence[str | ImageData] | None = None,
         *,
         in_args: InT | None = None,
+        call_id: str,
         ctx: RunContext[CtxT],
     ) -> UserMessage | None:
         if chat_inputs is not None:
@@ -116,7 +130,9 @@ class PromptBuilder(AutoInstanceAttributesMixin, Generic[InT, CtxT]):
             return UserMessage.from_content_parts(chat_inputs, name=self._agent_name)
 
         return UserMessage(
-            content=self._build_input_content(in_args=in_args, ctx=ctx),
+            content=self._build_input_content(
+                in_args=in_args, ctx=ctx, call_id=call_id
+            ),
             name=self._agent_name,
         )
 

grasp_agents/typing/tool.py

@@ -64,20 +64,22 @@ class BaseTool(
         self,
         inp: _InT,
         *,
-        call_id: str | None = None,
         ctx: RunContext[CtxT] | None = None,
+        call_id: str | None = None,
     ) -> _OutT_co:
         pass
 
     async def __call__(
         self,
         *,
-        call_id: str | None = None,
         ctx: RunContext[CtxT] | None = None,
+        call_id: str | None = None,
         **kwargs: Any,
     ) -> _OutT_co:
+        # NOTE: validation is probably redundant here when tool inputs have been
+        # validated by the LLM already
         input_args = TypeAdapter(self._in_type).validate_python(kwargs)
-        output = await self.run(input_args, call_id=call_id, ctx=ctx)
+        output = await self.run(input_args, ctx=ctx, call_id=call_id)
 
         return TypeAdapter(self._out_type).validate_python(output)
 

{grasp_agents-0.5.10.dist-info → grasp_agents-0.5.11.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: grasp_agents
-Version: 0.5.10
+Version: 0.5.11
 Summary: Grasp Agents Library
 License-File: LICENSE.md
 Requires-Python: <4,>=3.11.4
@@ -166,9 +166,7 @@ class AskStudentTool(BaseTool[TeacherQuestion, StudentReply, Any]):
     name: str = "ask_student"
     description: str = ask_student_tool_description
 
-    async def run(
-        self, inp: TeacherQuestion, ctx: RunContext[Any] | None = None
-    ) -> StudentReply:
+    async def run(self, inp: TeacherQuestion, **kwargs: Any) -> StudentReply:
         return input(inp.question)
 
 
@@ -180,7 +178,8 @@ teacher = LLMAgent[None, Problem, None](
     name="teacher",
     llm=LiteLLM(
         model_name="gpt-4.1",
-        llm_settings=LiteLLMSettings(temperature=0.5),
+        # model_name="claude-sonnet-4-20250514",
+        # llm_settings=LiteLLMSettings(reasoning_effort="low"),
     ),
     tools=[AskStudentTool()],
     react_mode=True,

{grasp_agents-0.5.10.dist-info → grasp_agents-0.5.11.dist-info}/RECORD

@@ -1,19 +1,19 @@
 grasp_agents/__init__.py,sha256=Z3a_j2Etiap9H6lvE8-PQP_OIGMUcHNPeJAJO12B8kY,1031
-grasp_agents/cloud_llm.py,sha256=_5iQAYL7w49-5oLgx0eycFJRoGSwh38tuFRWMkCxc0Q,13183
+grasp_agents/cloud_llm.py,sha256=vwI6gpLOsFqN4KtaTOo75xw8t7uRtdVrYGjopEDmQBw,13091
 grasp_agents/costs_dict.yaml,sha256=2MFNWtkv5W5WSCcv1Cj13B1iQLVv5Ot9pS_KW2Gu2DA,2510
 grasp_agents/errors.py,sha256=K-22TCM1Klhsej47Rg5eTqnGiGPaXgKOpdOZZ7cPipw,4633
 grasp_agents/generics_utils.py,sha256=5Pw3I9dlnKC2VGqYKC4ZZUO3Z_vTNT-NPFovNfPkl6I,6542
 grasp_agents/grasp_logging.py,sha256=H1GYhXdQvVkmauFDZ-KDwvVmPQHZUUm9sRqX_ObK2xI,1111
 grasp_agents/http_client.py,sha256=Es8NXGDkp4Nem7g24-jW0KFGA9Hp_o2Cv3cOvjup-iU,859
 grasp_agents/llm.py,sha256=IeV2QpR4AldVP3THzSETEnsaDx3DYz5HM6dkikSpy4o,10684
-grasp_agents/llm_agent.py,sha256=bkd_MvhQOJsH8CXJO-r2fexH5272bCkNWm8OOOSEf1I,13546
-grasp_agents/llm_agent_memory.py,sha256=uxyJ1jq0oUgxIh7ysRcorGy1Q1WHosLBvWqjvLT2AUg,1834
-grasp_agents/llm_policy_executor.py,sha256=JOAnIHZX-_oX1TcgCqRvIFQ6OuPQ4TypSvAxtT4ZGeg,17916
+grasp_agents/llm_agent.py,sha256=F_ou0pfdztqZzd2yU1jZZZVzcyhsLXfE_i0c4y2fZIQ,14123
+grasp_agents/llm_agent_memory.py,sha256=XmOT2G8RG5AHd0LR3WuK7VbD-KFFfThmJnuZK2iU3Fs,1856
+grasp_agents/llm_policy_executor.py,sha256=r0UxwjnVzTBQqLlwvZZ_JL0wl6ZebCgxkcz6I4GdmrM,18136
 grasp_agents/memory.py,sha256=keHuNEZNSxHT9FKpMohHOCNi7UAz_oRIc91IQEuzaWE,1162
 grasp_agents/packet.py,sha256=EmE-W4ZSMVZoqClECGFe7OGqrT4FSJ8IVGICrdjtdEY,1462
 grasp_agents/packet_pool.py,sha256=AF7ZMYY1U6ppNLEn6o0R8QXyWmcLQGcju7_TYQpAudg,4443
 grasp_agents/printer.py,sha256=wVNCaR9mbFKyzYdT8YpYD1JQqRqHdLtdfiZrwYxaM6Y,11132
-grasp_agents/prompt_builder.py,sha256=m8EZBFUOjK63ym9z-ix0R1R8V9hocB0mCNB6EKyeVmk,5829
+grasp_agents/prompt_builder.py,sha256=wNPphkW8RL8501jV4Z7ncsN_sxBDR9Ax7eILLHr-OYg,6110
 grasp_agents/run_context.py,sha256=7qVs0T5rLvINmtlXqOoyy2Hu9xPzuFDbcVR6R93NF-0,951
 grasp_agents/runner.py,sha256=JL2wSKahbPYVd56NRB09cwco43sjhZPI4XYFCZyOXOA,5173
 grasp_agents/usage_tracker.py,sha256=ZQfVUUpG0C89hyPWT_JgXnjQOxoYmumcQ9t-aCfcMo8,3561
@@ -30,11 +30,11 @@ grasp_agents/openai/completion_converters.py,sha256=UlDeQSl0AEFUS-QI5e8rrjfmXZoj
 grasp_agents/openai/content_converters.py,sha256=sMsZhoatuL_8t0IdVaGWIVZLB4nyi1ajD61GewQmeY4,2503
 grasp_agents/openai/converters.py,sha256=RKOfMbIJmfFQ7ot0RGR6wrdMbR6_L7PB0UZwxwgM88g,4691
 grasp_agents/openai/message_converters.py,sha256=fhSN81uK51EGbLyM2-f0MvPX_UBrMy7SF3JQPo-dkXg,4686
-grasp_agents/openai/openai_llm.py,sha256=mC73ImgRLd56DLzCbvFvtMyf7wjL6Hhelc7e-UNP2Q0,9715
+grasp_agents/openai/openai_llm.py,sha256=QjxrZ4fM_FX3ncBjehUjWPCCiI62u_W2XDi7nth1WrY,9737
 grasp_agents/openai/tool_converters.py,sha256=rNH5t2Wir9nuy8Ei0jaxNuzDaXGqTLmLz3VyrnJhyn0,1196
-grasp_agents/processors/base_processor.py,sha256=rQncbE6Cih3PoaGEkYQtTYz2PImP__tOaDb7TYLIoJY,10835
-grasp_agents/processors/parallel_processor.py,sha256=lpJSSmyUzGGDedqPdQ2ygBXwNK8MiEi7VZo-TCzJFeE,7945
-grasp_agents/processors/processor.py,sha256=DbitqHMQJCMCMTCvSQZQuamjEEC6e7RlN_L-Z4gA8vA,5283
+grasp_agents/processors/base_processor.py,sha256=BQ2k8dJY0jTMmidXZdK7JLO2YIQkmkp5boF1fT1o6uQ,10838
+grasp_agents/processors/parallel_processor.py,sha256=BOXRlPaZ-hooz0hHctqiW_5ldR-yDPYjFxuP7fAbZCI,7911
+grasp_agents/processors/processor.py,sha256=35MtYKrKtCZZMhV-U1DXBXtCNbCvZGaiiXo_5a3tI6s,5249
 grasp_agents/rate_limiting/__init__.py,sha256=KRgtF_E7R3YfA2cpYcFcZ7wycV0pWVJ0xRQC7YhiIEQ,158
 grasp_agents/rate_limiting/rate_limiter_chunked.py,sha256=BPgkUXvhmZhTpZs2T6uujNFuxH_kYHiISuf6_-eNhUc,5544
 grasp_agents/rate_limiting/types.py,sha256=PbnNhEAcYedQdIpPJWud8HUVcxa_xZS2RDZu4c5jr40,1003
@@ -47,12 +47,12 @@ grasp_agents/typing/converters.py,sha256=VrsqjuC_1IMj9rTOAMPBJ1N0hHY3Z9fx0zySo4Z
 grasp_agents/typing/events.py,sha256=vFq6qRGofY8NuxOG9ZIN2_CnhAqsAodYLD4b4KtAq2U,12620
 grasp_agents/typing/io.py,sha256=MGEoUjAwKH1AHYglFkKNpHiielw-NFf13Epg3B4Q7Iw,139
 grasp_agents/typing/message.py,sha256=o7bN84AgrC5Fm3Wx20gqL9ArAMcEtYvnHnXbb04ngCs,3224
-grasp_agents/typing/tool.py,sha256=ivUq7FIYq1St3GNzbl_1VDbnKOZkTjQB5dYAgj6GPX0,1913
+grasp_agents/typing/tool.py,sha256=qwC5baRratcyJWLMQ923IMGHH1hmj9eUtYLnNBcbwUU,2033
 grasp_agents/workflow/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
 grasp_agents/workflow/looped_workflow.py,sha256=WHp9O3Za2sBVfY_BLOdvPvtY20XsjZQaWSO2-oAFvOY,6806
 grasp_agents/workflow/sequential_workflow.py,sha256=e3BIWzy_2novmEWNwIteyMbrzvl1-evHrTBE3r3SpU8,3648
 grasp_agents/workflow/workflow_processor.py,sha256=DwHz70UOTp9dkbtzH9KE5LkGcT1RdHV7Hdiby0Bu9tw,3535
-grasp_agents-0.5.10.dist-info/METADATA,sha256=tprOY_Kg0RWdYp5PLA2iQuVcWq5BOLga4uPoJB4UdLI,6998
-grasp_agents-0.5.10.dist-info/WHEEL,sha256=qtCwoSJWgHk21S1Kb4ihdzI2rlJ1ZKaIurTj_ngOhyQ,87
-grasp_agents-0.5.10.dist-info/licenses/LICENSE.md,sha256=-nNNdWqGB8gJ2O-peFQ2Irshv5tW5pHKyTcYkwvH7CE,1201
-grasp_agents-0.5.10.dist-info/RECORD,,
+grasp_agents-0.5.11.dist-info/METADATA,sha256=BkVyEN63RzGsCIJCnm5S38EI2ua9NcbPmr3lRCmWPGs,7021
+grasp_agents-0.5.11.dist-info/WHEEL,sha256=qtCwoSJWgHk21S1Kb4ihdzI2rlJ1ZKaIurTj_ngOhyQ,87
+grasp_agents-0.5.11.dist-info/licenses/LICENSE.md,sha256=-nNNdWqGB8gJ2O-peFQ2Irshv5tW5pHKyTcYkwvH7CE,1201
+grasp_agents-0.5.11.dist-info/RECORD,,