diff --git a/autogpt/api_manager.py b/autogpt/api_manager.py
index ace64fbd..52b94ff7 100644
--- a/autogpt/api_manager.py
+++ b/autogpt/api_manager.py
@@ -1,22 +1,19 @@
-from typing import List
+from __future__ import annotations
 
 import openai
 
 from autogpt.config import Config
 from autogpt.logs import logger
 from autogpt.modelsinfo import COSTS
-
-cfg = Config()
-print_total_cost = cfg.debug_mode
+from autogpt.singleton import Singleton
 
 
-class ApiManager:
-    def __init__(self, debug=False):
+class ApiManager(metaclass=Singleton):
+    def __init__(self):
         self.total_prompt_tokens = 0
         self.total_completion_tokens = 0
         self.total_cost = 0
         self.total_budget = 0
-        self.debug = debug
 
     def reset(self):
         self.total_prompt_tokens = 0
@@ -28,7 +25,7 @@ class ApiManager:
         self,
         messages: list,  # type: ignore
         model: str | None = None,
-        temperature: float = cfg.temperature,
+        temperature: float = None,
         max_tokens: int | None = None,
         deployment_id=None,
     ) -> str:
@@ -42,6 +39,9 @@ class ApiManager:
         Returns:
         str: The AI's response.
         """
+        cfg = Config()
+        if temperature is None:
+            temperature = cfg.temperature
         if deployment_id is not None:
             response = openai.ChatCompletion.create(
                 deployment_id=deployment_id,
@@ -59,8 +59,7 @@ class ApiManager:
                 max_tokens=max_tokens,
                 api_key=cfg.openai_api_key,
             )
-        if self.debug:
-            logger.debug(f"Response: {response}")
+        logger.debug(f"Response: {response}")
         prompt_tokens = response.usage.prompt_tokens
         completion_tokens = response.usage.completion_tokens
         self.update_cost(prompt_tokens, completion_tokens, model)
@@ -81,8 +80,7 @@ class ApiManager:
             prompt_tokens * COSTS[model]["prompt"]
             + completion_tokens * COSTS[model]["completion"]
         ) / 1000
-        if print_total_cost:
-            print(f"Total running cost: ${self.total_cost:.3f}")
+        logger.debug(f"Total running cost: ${self.total_cost:.3f}")
 
     def set_total_budget(self, total_budget):
         """
@@ -128,6 +126,3 @@ class ApiManager:
         float: The total budget for API calls.
         """
         return self.total_budget
-
-
-api_manager = ApiManager(cfg.debug_mode)
diff --git a/autogpt/chat.py b/autogpt/chat.py
index 21eab6a0..4b906a00 100644
--- a/autogpt/chat.py
+++ b/autogpt/chat.py
@@ -3,7 +3,7 @@ import time
 from openai.error import RateLimitError
 
 from autogpt import token_counter
-from autogpt.api_manager import api_manager
+from autogpt.api_manager import ApiManager
 from autogpt.config import Config
 from autogpt.llm_utils import create_chat_completion
 from autogpt.logs import logger
@@ -134,6 +134,7 @@ def chat_with_ai(
                 # Move to the next most recent message in the full message history
                 next_message_to_add_index -= 1
 
+            api_manager = ApiManager()
             # inform the AI about its remaining budget (if it has one)
             if api_manager.get_total_budget() > 0.0:
                 remaining_budget = (
diff --git a/autogpt/llm_utils.py b/autogpt/llm_utils.py
index 212928ca..a98b12a3 100644
--- a/autogpt/llm_utils.py
+++ b/autogpt/llm_utils.py
@@ -8,7 +8,7 @@ import openai
 from colorama import Fore, Style
 from openai.error import APIError, RateLimitError, Timeout
 
-from autogpt.api_manager import api_manager
+from autogpt.api_manager import ApiManager
 from autogpt.config import Config
 from autogpt.logs import logger
 from autogpt.types.openai import Message
@@ -147,6 +147,7 @@ def create_chat_completion(
             )
             if message is not None:
                 return message
+    api_manager = ApiManager()
     response = None
     for attempt in range(num_retries):
         backoff = 2 ** (attempt + 2)
@@ -228,6 +229,7 @@ def get_ada_embedding(text: str) -> List[float]:
         kwargs = {"model": model}
 
     embedding = create_embedding(text, **kwargs)
+    api_manager = ApiManager()
     api_manager.update_cost(
         prompt_tokens=embedding.usage.prompt_tokens,
         completion_tokens=0,
diff --git a/autogpt/memory/base.py b/autogpt/memory/base.py
index a6c92db6..d99cadde 100644
--- a/autogpt/memory/base.py
+++ b/autogpt/memory/base.py
@@ -1,11 +1,8 @@
 """Base class for memory providers."""
 import abc
 
-from autogpt.config import Config
 from autogpt.singleton import AbstractSingleton
 
-cfg = Config()
-
 
 class MemoryProviderSingleton(AbstractSingleton):
     @abc.abstractmethod
diff --git a/autogpt/prompts/prompt.py b/autogpt/prompts/prompt.py
index cc06fabb..68d8a336 100644
--- a/autogpt/prompts/prompt.py
+++ b/autogpt/prompts/prompt.py
@@ -1,6 +1,6 @@
 from colorama import Fore
 
-from autogpt.api_manager import api_manager
+from autogpt.api_manager import ApiManager
 from autogpt.config.ai_config import AIConfig
 from autogpt.config.config import Config
 from autogpt.logs import logger
@@ -115,6 +115,7 @@ Continue (y/n): """
         config.save(CFG.ai_settings_file)
 
     # set the total api budget
+    api_manager = ApiManager()
     api_manager.set_total_budget(config.api_budget)
 
     # Agent Created, print message
diff --git a/tests/conftest.py b/tests/conftest.py
index 9ab49189..745f8b5d 100644
--- a/tests/conftest.py
+++ b/tests/conftest.py
@@ -3,7 +3,6 @@ from pathlib import Path
 import pytest
 
 from autogpt.api_manager import ApiManager
-from autogpt.api_manager import api_manager as api_manager_
 from autogpt.config import Config
 from autogpt.workspace import Workspace
 
@@ -32,7 +31,6 @@ def config(workspace: Workspace) -> Config:
 
 @pytest.fixture()
 def api_manager() -> ApiManager:
-    old_attrs = api_manager_.__dict__.copy()
-    api_manager_.reset()
-    yield api_manager_
-    api_manager_.__dict__.update(old_attrs)
+    if ApiManager in ApiManager._instances:
+        del ApiManager._instances[ApiManager]
+    return ApiManager()
diff --git a/tests/unit/test_commands.py b/tests/unit/test_commands.py
index a749cc6d..8596e1c9 100644
--- a/tests/unit/test_commands.py
+++ b/tests/unit/test_commands.py
@@ -3,8 +3,7 @@ from unittest.mock import MagicMock, patch
 
 import pytest
 
-import autogpt.agent.agent_manager as agent_manager
-from autogpt.app import execute_command, list_agents, start_agent
+from autogpt.app import list_agents, start_agent
 from tests.utils import requires_api_key
 
 
@@ -14,9 +13,11 @@ def test_make_agent() -> None:
     """Test that an agent can be created"""
     # Use the mock agent manager to avoid creating a real agent
     with patch("openai.ChatCompletion.create") as mock:
-        obj = MagicMock()
-        obj.response.choices[0].messages[0].content = "Test message"
-        mock.return_value = obj
+        response = MagicMock()
+        response.choices[0].messages[0].content = "Test message"
+        response.usage.prompt_tokens = 1
+        response.usage.completion_tokens = 1
+        mock.return_value = response
         start_agent("Test Agent", "chat", "Hello, how are you?", "gpt-3.5-turbo")
         agents = list_agents()
         assert "List of agents:\n0: chat" == agents