diff --git a/litellm/integrations/helicone.py b/litellm/integrations/helicone.py
index 6b3d61965..9e74b246f 100644
--- a/litellm/integrations/helicone.py
+++ b/litellm/integrations/helicone.py
@@ -2,7 +2,6 @@
 #    On success, logs events to Helicone
 import dotenv, os
 import requests
-from anthropic import HUMAN_PROMPT, AI_PROMPT
 dotenv.load_dotenv() # Loading env variables using dotenv
 import traceback
 class HeliconeLogger:
@@ -14,6 +13,7 @@ class HeliconeLogger:
         self.key = os.getenv('HELICONE_API_KEY')
 
     def claude_mapping(self, model, messages, response_obj):
+        from anthropic import HUMAN_PROMPT, AI_PROMPT
         prompt = f"{HUMAN_PROMPT}" 
         for message in messages:
             if "role" in message:
diff --git a/litellm/main.py b/litellm/main.py
index 7803de2a9..ea4d43a63 100644
--- a/litellm/main.py
+++ b/litellm/main.py
@@ -1,6 +1,5 @@
-import os, openai, cohere, replicate, sys
+import os, openai, sys
 from typing import Any
-from anthropic import Anthropic, HUMAN_PROMPT, AI_PROMPT
 from functools import partial
 import dotenv, traceback, random, asyncio, time
 from copy import deepcopy
@@ -13,7 +12,7 @@ from tenacity import (
     stop_after_attempt,
     wait_random_exponential,
 )  # for exponential backoff
-from litellm.utils import get_secret
+from litellm.utils import get_secret, install_and_import
 ####### ENVIRONMENT VARIABLES ###################
 dotenv.load_dotenv() # Loading env variables using dotenv
 
@@ -28,9 +27,7 @@ new_response = {
           }
         ]
       }
-# TODO move this to utils.py
 # TODO add translations
-# TODO see if this worked - model_name == krrish
 ####### COMPLETION ENDPOINTS ################
 #############################################
 async def acompletion(*args, **kwargs):
@@ -68,6 +65,7 @@ def completion(
       openai.api_type = "azure"
       openai.api_base = litellm.api_base if litellm.api_base is not None else get_secret("AZURE_API_BASE")
       openai.api_version = litellm.api_version if litellm.api_version is not None else get_secret("AZURE_API_VERSION")
+      # set key
       if api_key:
           openai.api_key = api_key
       elif litellm.azure_key:
@@ -92,6 +90,7 @@ def completion(
         )
     elif model in litellm.open_ai_chat_completion_models:
       openai.api_type = "openai"
+      # note: if a user sets a custom base - we should ensure this works
       openai.api_base = litellm.api_base if litellm.api_base is not None else "https://api.openai.com/v1"
       openai.api_version = None
       if litellm.organization:
@@ -155,6 +154,8 @@ def completion(
       model_response["usage"] = response["usage"]
       response = model_response
     elif "replicate" in model:
+      # import replicate/if it fails then pip install replicate
+      install_and_import("replicate")
       # replicate defaults to os.environ.get("REPLICATE_API_TOKEN")
       # checking in case user set it to REPLICATE_API_KEY instead 
       if not get_secret("REPLICATE_API_TOKEN") and get_secret("REPLICATE_API_KEY"):
@@ -194,6 +195,10 @@ def completion(
         }
       response = model_response
     elif model in litellm.anthropic_models:
+      # import anthropic/if it fails then pip install anthropic
+      install_and_import("anthropic")
+      from anthropic import Anthropic, HUMAN_PROMPT, AI_PROMPT
+
       #anthropic defaults to os.environ.get("ANTHROPIC_API_KEY")
       if api_key:
          os.environ["ANTHROPIC_API_KEY"] = api_key
@@ -239,6 +244,8 @@ def completion(
         }
       response = model_response
     elif model in litellm.cohere_models:
+      # import cohere/if it fails then pip install cohere
+      install_and_import("cohere")
       if api_key:
         cohere_key = api_key
       elif litellm.cohere_key:
diff --git a/litellm/tests/test_completion.py b/litellm/tests/test_completion.py
index a4c151e5b..925483f32 100644
--- a/litellm/tests/test_completion.py
+++ b/litellm/tests/test_completion.py
@@ -7,8 +7,10 @@ sys.path.insert(0, os.path.abspath('../..'))  # Adds the parent directory to the
 import pytest
 import litellm
 from litellm import embedding, completion
+from infisical import InfisicalClient
 
 # litellm.set_verbose = True
+litellm.secret_manager_client = InfisicalClient(token=os.environ["INFISICAL_TOKEN"])
 
 user_message = "Hello, whats the weather in San Francisco??"
 messages = [{ "content": user_message,"role": "user"}]
@@ -16,6 +18,14 @@ messages = [{ "content": user_message,"role": "user"}]
 def logger_fn(user_model_dict):
     print(f"user_model_dict: {user_model_dict}")
 
+def test_completion_claude():
+    try:
+        response = completion(model="claude-instant-1", messages=messages, logger_fn=logger_fn)
+        # Add any assertions here to check the response
+        print(response)
+    except Exception as e:
+        pytest.fail(f"Error occurred: {e}")
+
 def test_completion_openai():
     try:
         response = completion(model="gpt-3.5-turbo", messages=messages)
@@ -84,14 +94,6 @@ def test_completion_azure():
     except Exception as e:
         pytest.fail(f"Error occurred: {e}")
 
-def test_completion_claude():
-    try:
-        response = completion(model="claude-instant-1", messages=messages, logger_fn=logger_fn)
-        # Add any assertions here to check the response
-        print(response)
-    except Exception as e:
-        pytest.fail(f"Error occurred: {e}")
-
 def test_completion_cohere():
     try:
         response = completion(model="command-nightly", messages=messages, max_tokens=500)
diff --git a/litellm/utils.py b/litellm/utils.py
index d32b94689..5b4820131 100644
--- a/litellm/utils.py
+++ b/litellm/utils.py
@@ -4,7 +4,6 @@ import subprocess, os
 import litellm, openai 
 import random, uuid, requests
 import datetime, time
-from anthropic import Anthropic
 import tiktoken
 encoding = tiktoken.get_encoding("cl100k_base")
 from .integrations.helicone import HeliconeLogger
@@ -34,6 +33,19 @@ def print_verbose(print_statement):
     if random.random() <= 0.3:
       print("Get help - https://discord.com/invite/wuPM9dRgDw")
 
+####### Package Import Handler ###################
+import importlib
+import subprocess
+def install_and_import(package):
+    try:
+        importlib.import_module(package)
+    except ImportError:
+        print(f"{package} is not installed. Installing...")
+        subprocess.call([sys.executable, '-m', 'pip', 'install', package])
+    finally:
+        globals()[package] = importlib.import_module(package)
+##################################################
+
 ####### LOGGING ###################
 #Logging function -> log the exact model details + what's being sent | Non-Blocking
 def logging(model=None, input=None, azure=False, additional_args={}, logger_fn=None, exception=None):
@@ -329,6 +341,8 @@ def prompt_token_calculator(model, messages):
   text = " ".join(message["content"] for message in messages)
   num_tokens = 0
   if "claude" in model:
+    install_and_import('anthropic')
+    from anthropic import Anthropic, HUMAN_PROMPT, AI_PROMPT
     anthropic = Anthropic()
     num_tokens = anthropic.count_tokens(text)
   else:
diff --git a/pyproject.toml b/pyproject.toml
index 9953d8fe4..20916e782 100644
--- a/pyproject.toml
+++ b/pyproject.toml
@@ -8,12 +8,11 @@ readme = "README.md"
 
 [tool.poetry.dependencies]
 python = "^3.8"
-openai = {extras = ["datalib"], version = "^0.27.8"}
-cohere = "^4.18.0"
+openai = "^0.27.8"
+
 pytest = "^7.4.0"
-pydantic = "^2.1.1"
-anthropic = "^0.3.7"
-replicate = "^0.10.0"
+
+
 python-dotenv = "^1.0.0"
 tenacity = "^8.0.1"
 tiktoken = "^0.4.0"
diff --git a/requirements.txt b/requirements.txt
index 87aa1e737..ba5e487e3 100644
--- a/requirements.txt
+++ b/requirements.txt
@@ -1,8 +1,5 @@
-pydantic
+# used by CI/CD testing
 openai
-cohere
-anthropic
-replicate
 pytest
 python-dotenv
 openai[datalib]