Bug fixes and improvements

agency_swarm/agents/agent.py

@@ -9,6 +9,7 @@
 from openai.lib._parsing._completions import type_to_response_format_param
 from openai.types.beta.assistant import ToolResources
 
+from agency_swarm.constants import DEFAULT_MODEL
 from agency_swarm.tools import (
     BaseTool,
     CodeInterpreter,
@@ -17,7 +18,6 @@
     ToolFactory,
 )
 from agency_swarm.tools.oai.FileSearch import FileSearchConfig
-from agency_swarm.util.constants import DEFAULT_MODEL
 from agency_swarm.util.oai import get_openai_client
 from agency_swarm.util.openapi import validate_openapi_spec
 from agency_swarm.util.shared_state import SharedState

agency_swarm/threads/thread.py

@@ -194,20 +194,21 @@ def get_completion(
             )
 
         # chat model start callback
-        chat_messages = (
-            [[HumanMessage(content=message)]] if isinstance(message, str) else []
-        )
-        if self.callback_handler and chat_messages:
-            self.callback_handler.on_chat_model_start(
-                serialized={"name": self._run.model},
-                messages=chat_messages,
-                run_id=self._run.id,
-                parent_run_id=chain_run_id,
-                metadata={
-                    "agent_name": self.agent.name,
-                    "recipient_agent_name": recipient_agent.name,
-                },
+        if self.callback_handler:
+            chat_messages = (
+                [[HumanMessage(content=message)]] if isinstance(message, str) else []
             )
+            if chat_messages:
+                self.callback_handler.on_chat_model_start(
+                    serialized={"name": self._run.model},
+                    messages=chat_messages,
+                    run_id=self._run.id,
+                    parent_run_id=chain_run_id,
+                    metadata={
+                        "agent_name": self.agent.name,
+                        "recipient_agent_name": recipient_agent.name,
+                    },
+                )
 
         try:
             error_attempts = 0

agency_swarm/util/tracking/__init__.py

@@ -5,7 +5,8 @@
 _lock = threading.Lock()
 
 
-SUPPORTED_TRACKERS = Literal["langfuse", "local"]
+SUPPORTED_TRACKERS = ["langfuse", "local"]
+SUPPORTED_TRACKERS_TYPE = Literal["langfuse", "local"]
 
 
 def get_callback_handler():
@@ -20,7 +21,7 @@ def set_callback_handler(handler: Callable):
         _callback_handler = handler()
 
 
-def init_tracking(tracker_name: SUPPORTED_TRACKERS, **kwargs):
+def init_tracking(tracker_name: SUPPORTED_TRACKERS_TYPE, **kwargs):
     if tracker_name not in SUPPORTED_TRACKERS:
         raise ValueError(f"Invalid tracker name: {tracker_name}")
 

agency_swarm/util/tracking/langchain_types.py

@@ -1,14 +1,33 @@
-from typing import TYPE_CHECKING, TypeVar
+from typing import TYPE_CHECKING, Any, Dict
+
+from pydantic import BaseModel
 
 if TYPE_CHECKING:
     from langchain.schema import AgentAction as LangchainAgentAction
     from langchain.schema import AgentFinish as LangchainAgentFinish
     from langchain.schema import HumanMessage as LangchainHumanMessage
 
-# Define TypeVars that can be either our placeholder or langchain types
-AgentAction = TypeVar("AgentAction", bound="LangchainAgentAction")
-AgentFinish = TypeVar("AgentFinish", bound="LangchainAgentFinish")
-HumanMessage = TypeVar("HumanMessage", bound="LangchainHumanMessage")
+
+# Create base classes that match langchain's structure
+class BaseAgentAction(BaseModel):
+    tool: str
+    tool_input: Dict[str, Any] | str
+    log: str
+
+
+class BaseAgentFinish(BaseModel):
+    return_values: Dict[str, Any]
+    log: str
+
+
+class BaseHumanMessage(BaseModel):
+    content: str
+
+
+# Initialize with our base implementations first
+AgentAction = BaseAgentAction
+AgentFinish = BaseAgentFinish
+HumanMessage = BaseHumanMessage
 
 
 def use_langchain_types():

agency_swarm/util/tracking/local_callback_handler.py

@@ -6,10 +6,8 @@
 
 from langchain.schema import AgentAction, AgentFinish, BaseMessage, Document, LLMResult
 
-from agency_swarm.util.tracking.callbacks import CallbackHandler
 
-
-class LocalCallbackHandler(CallbackHandler):
+class LocalCallbackHandler:
     def __init__(self, db_path: str = "usage.db"):
         self.conn = sqlite3.connect(db_path, check_same_thread=False)
         self.lock = threading.Lock()

agency_swarm/util/validators.py

@@ -3,7 +3,7 @@
 from openai import OpenAI
 from pydantic import BaseModel, Field
 
-from agency_swarm.util.constants import DEFAULT_MODEL_MINI
+from agency_swarm.constants import DEFAULT_MODEL_MINI
 from agency_swarm.util.oai import get_openai_client
 
 

pyproject.toml

@@ -25,7 +25,7 @@ dependencies = [
     "rich==13.7.1",
     "termcolor==2.4.0",
 ]
-requires-python = ">=3.10,<3.13"
+requires-python = ">=3.10"
 urls = { homepage = "https://github.com/VRSEN/agency-swarm" }
 
 [project.scripts]

requirements_test.txt

@@ -1,2 +1,3 @@
 langchain==0.3.13
+langchain-community==0.3.13
 pytest

tests/demos/demo_observability.py

@@ -1,16 +1,15 @@
 from dotenv import load_dotenv
 
-load_dotenv()
+from agency_swarm import Agency, Agent
+from agency_swarm.util import init_tracking
 
-from agency_swarm import Agency, Agent  # noqa
-from agency_swarm.util import init_tracking  # noqa
+load_dotenv()
 
 
 def main():
     # Set the tracker type
-    TRACKER = "local"
-    # To use Langfuse, uncomment the next line
-    # TRACKER = "langfuse"
+    # TRACKER = "local"
+    TRACKER = "langfuse"
 
     # Initialize tracking based on the selected tracker
     init_tracking(TRACKER)

tests/demos/streaming_demo.py

@@ -1,11 +1,9 @@
-import sys
 import time
 import unittest
 
 from agency_swarm import Agent, BaseTool
 from agency_swarm.agency.agency import Agency
-
-sys.path.insert(0, "../agency-swarm")
+from agency_swarm.constants import DEFAULT_MODEL_MINI
 
 
 class StreamingTest(unittest.TestCase):
@@ -19,12 +17,12 @@ def run(self):
         self.ceo = Agent(
             name="ceo",
             instructions="You are a CEO of an agency made for testing purposes.",
-            model="gpt-4o-mini",
+            model=DEFAULT_MODEL_MINI,
         )
         self.test_agent1 = Agent(
-            name="test_agent1", tools=[TestTool], model="gpt-4o-mini"
+            name="test_agent1", tools=[TestTool], model=DEFAULT_MODEL_MINI
         )
-        self.test_agent2 = Agent(name="test_agent2", model="gpt-4o-mini")
+        self.test_agent2 = Agent(name="test_agent2", model=DEFAULT_MODEL_MINI)
 
         self.agency = Agency(
             [

tests/test_communication.py

@@ -70,7 +70,10 @@ def test_send_message_double_recipient_error(self):
         ceo = Agent(
             name="CEO",
             description="Responsible for client communication, task planning and management.",
-            instructions="You are an agent for testing. Route request AT THE SAME TIME as instructed. If there is an error in a single request, please say 'error'. If there are errors in both requests, please say 'fatal'. do not output anything else.",
+            instructions="""You are an agent for testing. When asked to route requests AT THE SAME TIME:
+            1. If you detect multiple simultaneous routing requests, respond with 'error'
+            2. If you detect errors in all routing attempts, respond with 'fatal'
+            3. Do not output anything else besides these exact words.""",
         )
         test_agent = Agent(
             name="Test Agent1",
@@ -79,7 +82,7 @@ def test_send_message_double_recipient_error(self):
         )
         agency = Agency([ceo, [ceo, test_agent]], temperature=0)
         response = agency.get_completion(
-            "Please route me to customer support TWICE at the same time. I am testing something."
+            "Route me to customer support TWICE simultaneously (at the exact same time). This is a test of concurrent routing."
         )
         self.assertTrue("error" in response.lower(), agency.main_thread.thread_url)
         self.assertTrue("fatal" not in response.lower(), agency.main_thread.thread_url)