Platform for stateful agents: AI with advanced memory that can learn and self-improve over time.
feat: Enable dynamic toggling of tool choice in v3 agent loop for OpenAI [LET-4564] (#5042)
Matthew Zhoudf5c997da0ee
diff --git a/letta/agents/letta_agent_v3.py b/letta/agents/letta_agent_v3.pyindex 73436f7c7..063cb735e 100644--- a/letta/agents/letta_agent_v3.py+++ b/letta/agents/letta_agent_v3.py@@ -51,6 +51,10 @@ class LettaAgentV3(LettaAgentV2):* Support Gemini / OpenAI client"""+ def _initialize_state(self):+ super()._initialize_state()+ self._require_tool_call = False+@trace_methodasync def step(self,@@ -279,6 +283,15 @@ class LettaAgentV3(LettaAgentV2):try:self.last_function_response = _load_last_function_response(messages)valid_tools = await self._get_valid_tools()+ require_tool_call = self.tool_rules_solver.should_force_tool_call()++ if self._require_tool_call != require_tool_call:+ if require_tool_call:+ self.logger.info("switching to constrained mode (forcing tool call)")+ else:+ self.logger.info("switching to unconstrained mode (allowing non-tool responses)")+ self._require_tool_call = require_tool_call+approval_request, approval_response = _maybe_get_approval_messages(messages)if approval_request and approval_response:tool_call = approval_request.tool_calls[0]@@ -307,6 +320,7 @@ class LettaAgentV3(LettaAgentV2):llm_config=self.agent_state.llm_config,tools=valid_tools,force_tool_call=force_tool_call,+ requires_subsequent_tool_call=self._require_tool_call,)if dry_run:yield request_data@@ -590,8 +604,7 @@ class LettaAgentV3(LettaAgentV2):elif tool_call is None:# TODO could just hardcode the line here instead of calling the function...continue_stepping, heartbeat_reason, stop_reason = self._decide_continuation(- # agent_state=agent_state,- # request_heartbeat=False,+ agent_state=agent_state,tool_call_name=None,tool_rule_violated=False,tool_rules_solver=tool_rules_solver,@@ -705,8 +718,7 @@ class LettaAgentV3(LettaAgentV2):# 4. Decide whether to keep stepping (focal section simplified)continue_stepping, heartbeat_reason, stop_reason = self._decide_continuation(- # agent_state=agent_state,- # request_heartbeat=request_heartbeat,+ agent_state=agent_state,tool_call_name=tool_call_name,tool_rule_violated=tool_rule_violated,tool_rules_solver=tool_rules_solver,@@ -753,8 +765,7 @@ class LettaAgentV3(LettaAgentV2):@trace_methoddef _decide_continuation(self,- # agent_state: AgentState,- # request_heartbeat: bool,+ agent_state: AgentState,tool_call_name: Optional[str],tool_rule_violated: bool,tool_rules_solver: ToolRulesSolver,@@ -771,19 +782,14 @@ class LettaAgentV3(LettaAgentV2):2c. Called tool + tool rule violation (did not execute)"""+ continue_stepping = True # Default continuecontinuation_reason: str | None = Nonestop_reason: LettaStopReason | None = Noneif tool_call_name is None:# No tool call? End loopreturn False, None, LettaStopReason(stop_reason=StopReasonType.end_turn.value)-else:- # If we have a tool call, we continue stepping- return True, None, None-- # TODO support tool rules- # I think we can just uncomment the bellow?if tool_rule_violated:continue_stepping = Truecontinuation_reason = f"{NON_USER_MSG_PREFIX}Continuing: tool rule violation."@@ -791,8 +797,7 @@ class LettaAgentV3(LettaAgentV2):tool_rules_solver.register_tool_call(tool_call_name)if tool_rules_solver.is_terminal_tool(tool_call_name):- if continue_stepping:- stop_reason = LettaStopReason(stop_reason=StopReasonType.tool_rule.value)+ stop_reason = LettaStopReason(stop_reason=StopReasonType.tool_rule.value)continue_stepping = Falseelif tool_rules_solver.has_children_tools(tool_call_name):@@ -809,7 +814,7 @@ class LettaAgentV3(LettaAgentV2):stop_reason = LettaStopReason(stop_reason=StopReasonType.max_steps.value)else:uncalled = tool_rules_solver.get_uncalled_required_tools(available_tools=set([t.name for t in agent_state.tools]))- if not continue_stepping and uncalled:+ if uncalled:continue_stepping = Truecontinuation_reason = (f"{NON_USER_MSG_PREFIX}Continuing, user expects these tools: [{', '.join(uncalled)}] to be called still."@@ -817,7 +822,7 @@ class LettaAgentV3(LettaAgentV2):stop_reason = None # reset – we’re still going- return continue_stepping, continuation_reason, stop_reason+ return continue_stepping, continuation_reason, stop_reason@trace_methodasync def _get_valid_tools(self):diff --git a/letta/llm_api/anthropic_client.py b/letta/llm_api/anthropic_client.pyindex d9b33ccd5..7c8ebbb00 100644--- a/letta/llm_api/anthropic_client.py+++ b/letta/llm_api/anthropic_client.py@@ -183,6 +183,7 @@ class AnthropicClient(LLMClientBase):llm_config: LLMConfig,tools: Optional[List[dict]] = None,force_tool_call: Optional[str] = None,+ requires_subsequent_tool_call: bool = False,) -> dict:# TODO: This needs to get cleaned up. The logic here is pretty confusing.# TODO: I really want to get rid of prefixing, it's a recipe for disaster code maintenance wisediff --git a/letta/llm_api/bedrock_client.py b/letta/llm_api/bedrock_client.pyindex 4b0de30db..d2d5722c3 100644--- a/letta/llm_api/bedrock_client.py+++ b/letta/llm_api/bedrock_client.py@@ -70,8 +70,9 @@ class BedrockClient(AnthropicClient):llm_config: LLMConfig,tools: Optional[List[dict]] = None,force_tool_call: Optional[str] = None,+ requires_subsequent_tool_call: bool = False,) -> dict:- data = super().build_request_data(agent_type, messages, llm_config, tools, force_tool_call)+ data = super().build_request_data(agent_type, messages, llm_config, tools, force_tool_call, requires_subsequent_tool_call)# remove disallowed fieldsif "tool_choice" in data:del data["tool_choice"]["disable_parallel_tool_use"]diff --git a/letta/llm_api/deepseek_client.py b/letta/llm_api/deepseek_client.pyindex 3dd2d0df8..462468be6 100644--- a/letta/llm_api/deepseek_client.py+++ b/letta/llm_api/deepseek_client.py@@ -337,11 +337,12 @@ class DeepseekClient(OpenAIClient):llm_config: LLMConfig,tools: Optional[List[dict]] = None,force_tool_call: Optional[str] = None,+ requires_subsequent_tool_call: bool = False,) -> dict:# Override put_inner_thoughts_in_kwargs to False for DeepSeekllm_config.put_inner_thoughts_in_kwargs = False- data = super().build_request_data(agent_type, messages, llm_config, tools, force_tool_call)+ data = super().build_request_data(agent_type, messages, llm_config, tools, force_tool_call, requires_subsequent_tool_call)def add_functions_to_system_message(system_message: ChatMessage):system_message.content += f"<available functions> {''.join(json.dumps(f) for f in tools)} </available functions>"diff --git a/letta/llm_api/google_vertex_client.py b/letta/llm_api/google_vertex_client.pyindex 18be59b27..180b38ab3 100644--- a/letta/llm_api/google_vertex_client.py+++ b/letta/llm_api/google_vertex_client.py@@ -280,6 +280,7 @@ class GoogleVertexClient(LLMClientBase):llm_config: LLMConfig,tools: List[dict],force_tool_call: Optional[str] = None,+ requires_subsequent_tool_call: bool = False,) -> dict:"""Constructs a request object in the expected data format for this client.diff --git a/letta/llm_api/groq_client.py b/letta/llm_api/groq_client.pyindex 795449889..f9d3baae2 100644--- a/letta/llm_api/groq_client.py+++ b/letta/llm_api/groq_client.py@@ -29,8 +29,9 @@ class GroqClient(OpenAIClient):llm_config: LLMConfig,tools: Optional[List[dict]] = None,force_tool_call: Optional[str] = None,+ requires_subsequent_tool_call: bool = False,) -> dict:- data = super().build_request_data(agent_type, messages, llm_config, tools, force_tool_call)+ data = super().build_request_data(agent_type, messages, llm_config, tools, force_tool_call, requires_subsequent_tool_call)# Groq validation - these fields are not supported and will cause 400 errors# https://console.groq.com/docs/openaidiff --git a/letta/llm_api/llm_client_base.py b/letta/llm_api/llm_client_base.pyindex a8a3595b7..4a2d8980a 100644--- a/letta/llm_api/llm_client_base.py+++ b/letta/llm_api/llm_client_base.py@@ -127,6 +127,7 @@ class LLMClientBase:llm_config: LLMConfig,tools: List[dict],force_tool_call: Optional[str] = None,+ requires_subsequent_tool_call: bool = False,) -> dict:"""Constructs a request object in the expected data format for this client.diff --git a/letta/llm_api/openai_client.py b/letta/llm_api/openai_client.pyindex 0df9fb0f9..5e240c366 100644--- a/letta/llm_api/openai_client.py+++ b/letta/llm_api/openai_client.py@@ -206,6 +206,7 @@ class OpenAIClient(LLMClientBase):llm_config: LLMConfig,tools: Optional[List[dict]] = None, # Keep as dict for now as per base classforce_tool_call: Optional[str] = None,+ requires_subsequent_tool_call: bool = False,) -> dict:"""Constructs a request object in the expected data format for the OpenAI Responses API.@@ -224,14 +225,15 @@ class OpenAIClient(LLMClientBase):logger.warning(f"Model type not set in llm_config: {llm_config.model_dump_json(indent=4)}")model = None- # Default to auto, unless there's a forced tool call coming from above+ # Default to auto, unless there's a forced tool call coming from above or requires_subsequent_tool_call is Truetool_choice = Noneif tools: # only set tool_choice if tools exist- tool_choice = (- "auto"- if force_tool_call is None- else ToolFunctionChoice(type="function", function=ToolFunctionChoiceFunctionCall(name=force_tool_call))- )+ if force_tool_call is not None:+ tool_choice = {"type": "function", "name": force_tool_call}+ elif requires_subsequent_tool_call:+ tool_choice = "required"+ else:+ tool_choice = "auto"# Convert the tools from the ChatCompletions style to the Responses styleif tools:@@ -352,6 +354,7 @@ class OpenAIClient(LLMClientBase):llm_config: LLMConfig,tools: Optional[List[dict]] = None, # Keep as dict for now as per base classforce_tool_call: Optional[str] = None,+ requires_subsequent_tool_call: bool = False,) -> dict:"""Constructs a request object in the expected data format for the OpenAI API.@@ -364,6 +367,7 @@ class OpenAIClient(LLMClientBase):llm_config=llm_config,tools=tools,force_tool_call=force_tool_call,+ requires_subsequent_tool_call=requires_subsequent_tool_call,)if agent_type == AgentType.letta_v1_agent:@@ -407,15 +411,16 @@ class OpenAIClient(LLMClientBase):# TODO: This vllm checking is very brittle and is a patch at mosttool_choice = Noneif tools: # only set tool_choice if tools exist- if self.requires_auto_tool_choice(llm_config) or agent_type == AgentType.letta_v1_agent:+ if force_tool_call is not None:+ tool_choice = ToolFunctionChoice(type="function", function=ToolFunctionChoiceFunctionCall(name=force_tool_call))+ elif requires_subsequent_tool_call:+ tool_choice = "required"+ elif self.requires_auto_tool_choice(llm_config) or agent_type == AgentType.letta_v1_agent:tool_choice = "auto"else:# only set if tools is non-Nulltool_choice = "required"- if force_tool_call is not None:- tool_choice = ToolFunctionChoice(type="function", function=ToolFunctionChoiceFunctionCall(name=force_tool_call))-data = ChatCompletionRequest(model=model,messages=fill_image_content_in_messages(openai_message_list, messages),diff --git a/letta/llm_api/xai_client.py b/letta/llm_api/xai_client.pyindex 199edcc0b..f39dfbc90 100644--- a/letta/llm_api/xai_client.py+++ b/letta/llm_api/xai_client.py@@ -29,8 +29,9 @@ class XAIClient(OpenAIClient):llm_config: LLMConfig,tools: Optional[List[dict]] = None,force_tool_call: Optional[str] = None,+ requires_subsequent_tool_call: bool = False,) -> dict:- data = super().build_request_data(agent_type, messages, llm_config, tools, force_tool_call)+ data = super().build_request_data(agent_type, messages, llm_config, tools, force_tool_call, requires_subsequent_tool_call)# Specific bug for the mini models (as of Apr 14, 2025)# 400 - {'code': 'Client specified an invalid argument', 'error': 'Argument not supported on this model: presencePenalty'}diff --git a/tests/configs/llm_model_configs/openai-gpt-4.1.json b/tests/configs/llm_model_configs/openai-gpt-4.1.jsonnew file mode 100644index 000000000..0d073ae1c--- /dev/null+++ b/tests/configs/llm_model_configs/openai-gpt-4.1.json@@ -0,0 +1,7 @@+{+ "context_window": 1047576,+ "model": "gpt-4.1-2025-04-14",+ "model_endpoint_type": "openai",+ "model_endpoint": "https://api.openai.com/v1",+ "model_wrapper": null+}diff --git a/tests/integration_test_agent_tool_graph.py b/tests/integration_test_agent_tool_graph.pyindex e0e218ce6..162b0ee4f 100644--- a/tests/integration_test_agent_tool_graph.py+++ b/tests/integration_test_agent_tool_graph.py@@ -5,6 +5,7 @@ import uuidimport pytestfrom letta.agents.letta_agent_v2 import LettaAgentV2+from letta.agents.letta_agent_v3 import LettaAgentV3from letta.config import LettaConfigfrom letta.schemas.letta_message import ToolCallMessagefrom letta.schemas.message import MessageCreate@@ -12,7 +13,6 @@ from letta.schemas.run import Runfrom letta.schemas.tool_rule import ChildToolRule, ContinueToolRule, InitToolRule, RequiredBeforeExitToolRule, TerminalToolRulefrom letta.server.server import SyncServerfrom letta.services.run_manager import RunManager-from letta.services.telemetry_manager import NoopTelemetryManagerfrom tests.helpers.endpoints_helper import (assert_invoked_function_call,assert_invoked_send_message_with_keyword,@@ -25,7 +25,9 @@ from tests.utils import create_tool_from_func# Generate uuid for agent name for this examplenamespace = uuid.NAMESPACE_DNSagent_uuid = str(uuid.uuid5(namespace, "test_agent_tool_graph"))-config_file = "tests/configs/llm_model_configs/openai-gpt-4o.json"++OPENAI_CONFIG = "tests/configs/llm_model_configs/openai-gpt-4.1.json"+CLAUDE_SONNET_CONFIG = "tests/configs/llm_model_configs/claude-4-sonnet.json"@pytest.fixture()@@ -38,6 +40,12 @@ async def server():return server+def default_config_file():+ """Provides the default config file path for tests."""+ return OPENAI_CONFIG++@pytest.fixture(scope="function")async def first_secret_tool(server):def first_secret_word():@@ -241,7 +249,7 @@ async def default_user(server):async def run_agent_step(agent_state, input_messages, actor):"""Helper function to run agent step using LettaAgent directly instead of server.send_messages."""- agent_loop = LettaAgentV2(+ agent_loop = LettaAgentV3(agent_state=agent_state,actor=actor,)@@ -283,7 +291,7 @@ async def test_single_path_agent_tool_call_graph(]# Make agent state- agent_state = await setup_agent(server, config_file, agent_uuid=agent_uuid, tool_ids=[t.id for t in tools], tool_rules=tool_rules)+ agent_state = await setup_agent(server, OPENAI_CONFIG, agent_uuid=agent_uuid, tool_ids=[t.id for t in tools], tool_rules=tool_rules)response = await run_agent_step(agent_state=agent_state,input_messages=[MessageCreate(role="user", content="What is the fourth secret word?")],@@ -321,9 +329,8 @@ async def test_single_path_agent_tool_call_graph(@pytest.mark.parametrize("config_file",[- "tests/configs/llm_model_configs/claude-3-5-sonnet.json",- "tests/configs/llm_model_configs/openai-gpt-3.5-turbo.json",- "tests/configs/llm_model_configs/openai-gpt-4o.json",+ CLAUDE_SONNET_CONFIG,+ OPENAI_CONFIG,],)@pytest.mark.parametrize("init_tools_case", ["single", "multiple"])@@ -383,13 +390,12 @@ async def test_claude_initial_tool_rule_enforced(TerminalToolRule(tool_name=second_secret_tool.name),]tools = [first_secret_tool, second_secret_tool]- anthropic_config_file = "tests/configs/llm_model_configs/claude-3-5-sonnet.json"for i in range(3):agent_uuid = str(uuid.uuid4())agent_state = await setup_agent(server,- anthropic_config_file,+ CLAUDE_SONNET_CONFIG,agent_uuid=agent_uuid,tool_ids=[t.id for t in tools],tool_rules=tool_rules,@@ -426,8 +432,7 @@ async def test_claude_initial_tool_rule_enforced(@pytest.mark.parametrize("config_file",[- "tests/configs/llm_model_configs/claude-3-5-sonnet.json",- "tests/configs/llm_model_configs/openai-gpt-4o.json",+ OPENAI_CONFIG,],)@pytest.mark.asyncio@@ -508,13 +513,12 @@ async def test_init_tool_rule_always_fails(include_base_tools,):"""Test behavior when InitToolRule invokes a tool that always fails."""- config_file = "tests/configs/llm_model_configs/claude-3-5-sonnet.json"agent_uuid = str(uuid.uuid4())tool_rule = InitToolRule(tool_name=auto_error_tool.name)agent_state = await setup_agent(server,- config_file,+ OPENAI_CONFIG,agent_uuid=agent_uuid,tool_ids=[auto_error_tool.id],tool_rules=[tool_rule],@@ -535,7 +539,6 @@ async def test_init_tool_rule_always_fails(@pytest.mark.asyncioasync def test_continue_tool_rule(server, default_user):"""Test the continue tool rule by forcing send_message to loop before ending with core_memory_append."""- config_file = "tests/configs/llm_model_configs/claude-3-5-sonnet.json"agent_uuid = str(uuid.uuid4())tools = [@@ -551,7 +554,7 @@ async def test_continue_tool_rule(server, default_user):agent_state = await setup_agent(server,- config_file,+ CLAUDE_SONNET_CONFIG,agent_uuid,tool_ids=tool_ids,tool_rules=tool_rules,@@ -618,7 +621,7 @@ async def test_continue_tool_rule(server, default_user):# ]# tools = [flip_coin_tool, reveal_secret]#-# config_file = "tests/configs/llm_model_configs/claude-3-5-sonnet.json"+# config_file = CLAUDE_SONNET_CONFIG# agent_state = await setup_agent(client, config_file, agent_uuid=agent_uuid, tool_ids=[t.id for t in tools], tool_rules=tool_rules)# response = client.user_message(agent_id=agent_state.id, message="flip a coin until you get the secret word")#@@ -824,7 +827,6 @@ async def test_continue_tool_rule(server, default_user):async def test_single_required_before_exit_tool(server, disable_e2b_api_key, save_data_tool, default_user):"""Test that agent is forced to call a single required-before-exit tool before ending."""agent_name = "required_exit_single_tool_agent"- config_file = "tests/configs/llm_model_configs/openai-gpt-4o.json"# Set up tools and rulestools = [save_data_tool]@@ -835,7 +837,7 @@ async def test_single_required_before_exit_tool(server, disable_e2b_api_key, sav]# Create agent- agent_state = await setup_agent(server, config_file, agent_uuid=agent_name, tool_ids=[t.id for t in tools], tool_rules=tool_rules)+ agent_state = await setup_agent(server, OPENAI_CONFIG, agent_uuid=agent_name, tool_ids=[t.id for t in tools], tool_rules=tool_rules)# Send message that would normally cause exitresponse = await run_agent_step(@@ -866,7 +868,6 @@ async def test_single_required_before_exit_tool(server, disable_e2b_api_key, savasync def test_multiple_required_before_exit_tools(server, disable_e2b_api_key, save_data_tool, cleanup_temp_files_tool, default_user):"""Test that agent calls all required-before-exit tools before ending."""agent_name = "required_exit_multi_tool_agent"- config_file = "tests/configs/llm_model_configs/openai-gpt-4o.json"# Set up tools and rulestools = [save_data_tool, cleanup_temp_files_tool]@@ -878,7 +879,7 @@ async def test_multiple_required_before_exit_tools(server, disable_e2b_api_key,]# Create agent- agent_state = await setup_agent(server, config_file, agent_uuid=agent_name, tool_ids=[t.id for t in tools], tool_rules=tool_rules)+ agent_state = await setup_agent(server, OPENAI_CONFIG, agent_uuid=agent_name, tool_ids=[t.id for t in tools], tool_rules=tool_rules)# Send message that would normally cause exitresponse = await run_agent_step(@@ -911,7 +912,6 @@ async def test_multiple_required_before_exit_tools(server, disable_e2b_api_key,async def test_required_before_exit_with_other_rules(server, disable_e2b_api_key, first_secret_tool, save_data_tool, default_user):"""Test required-before-exit rules work alongside other tool rules."""agent_name = "required_exit_with_rules_agent"- config_file = "tests/configs/llm_model_configs/openai-gpt-4o.json"# Set up tools and rules - combine with child tool rulestools = [first_secret_tool, save_data_tool]@@ -923,7 +923,7 @@ async def test_required_before_exit_with_other_rules(server, disable_e2b_api_key]# Create agent- agent_state = await setup_agent(server, config_file, agent_uuid=agent_name, tool_ids=[t.id for t in tools], tool_rules=tool_rules)+ agent_state = await setup_agent(server, OPENAI_CONFIG, agent_uuid=agent_name, tool_ids=[t.id for t in tools], tool_rules=tool_rules)# Send message that would trigger tool flowresponse = await run_agent_step(@@ -956,7 +956,6 @@ async def test_required_before_exit_with_other_rules(server, disable_e2b_api_keyasync def test_required_tools_called_during_normal_flow(server, disable_e2b_api_key, save_data_tool, default_user):"""Test that agent can exit normally when required tools are called during regular operation."""agent_name = "required_exit_normal_flow_agent"- config_file = "tests/configs/llm_model_configs/openai-gpt-4o.json"# Set up tools and rulestools = [save_data_tool]@@ -967,7 +966,7 @@ async def test_required_tools_called_during_normal_flow(server, disable_e2b_api_]# Create agent- agent_state = await setup_agent(server, config_file, agent_uuid=agent_name, tool_ids=[t.id for t in tools], tool_rules=tool_rules)+ agent_state = await setup_agent(server, OPENAI_CONFIG, agent_uuid=agent_name, tool_ids=[t.id for t in tools], tool_rules=tool_rules)# Send message that explicitly mentions calling the required toolresponse = await run_agent_step(@@ -990,51 +989,3 @@ async def test_required_tools_called_during_normal_flow(server, disable_e2b_api_assert len(send_message_calls) == 1, "Should call send_message exactly once"print(f"✓ Agent '{agent_name}' exited cleanly after calling required tool normally")---async def test_terminal_tool_rule_send_message_request_heartbeat_false(server, disable_e2b_api_key, default_user):- """Test that when there's a terminal tool rule on send_message, the tool call has request_heartbeat=False."""- agent_name = "terminal_send_message_heartbeat_test"- config_file = "tests/configs/llm_model_configs/openai-gpt-4o.json"-- # Set up tool rules with terminal rule on send_message- tool_rules = [- TerminalToolRule(tool_name="send_message"),- ]-- # Create agent- agent_state = await setup_agent(server, config_file, agent_uuid=agent_name, tool_ids=[], tool_rules=tool_rules)-- # Send message that should trigger send_message tool call- response = await run_agent_step(- agent_state=agent_state,- input_messages=[MessageCreate(role="user", content="Please send me a simple message.")],- actor=default_user,- )-- # Assertions- assert_sanity_checks(response)- assert_invoked_function_call(response.messages, "send_message")-- # Find the send_message tool call and check request_heartbeat is False- send_message_call = None- for message in response.messages:- if isinstance(message, ToolCallMessage) and message.tool_call.name == "send_message":- send_message_call = message- break-- assert send_message_call is not None, "send_message tool call should be found"-- # Parse the arguments and check request_heartbeat- try:- arguments = json.loads(send_message_call.tool_call.arguments)- assert "request_heartbeat" in arguments, "request_heartbeat should be present in send_message arguments"- assert arguments["request_heartbeat"] is False, "request_heartbeat should be False for terminal tool rule"-- print(f"✓ Agent '{agent_name}' correctly set request_heartbeat=False for terminal send_message")- except json.JSONDecodeError:- pytest.fail("Failed to parse tool call arguments as JSON")- finally:- await cleanup_async(server=server, agent_uuid=agent_name, actor=default_user)