MemGPT AutoGen with Local LLM Error

[naseerfaheem]

Describe the bug
When trying to run examples from the autogen directory with local LLMs (WebUI and LMstudio), I get the OpenAI API error:
File "/home/user1/Documents/DataScience/installs/MemGPT/memgpt/autogen/examples/agent_docs.py", line 148, in
memgpt_agent = create_memgpt_autogen_agent_from_config(
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/home/user1/.conda/envs/memgpt/lib/python3.11/site-packages/memgpt/autogen/memgpt_agent.py", line 95, in create_memgpt_autogen_agent_from_config
autogen_memgpt_agent = create_autogen_memgpt_agent(
^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/home/user1/.conda/envs/memgpt/lib/python3.11/site-packages/memgpt/autogen/memgpt_agent.py", line 184, in create_autogen_memgpt_agent
autogen_memgpt_agent = MemGPTAgent(
^^^^^^^^^^^^
File "/home/user1/.conda/envs/memgpt/lib/python3.11/site-packages/memgpt/autogen/memgpt_agent.py", line 204, in init
super().init(name)
File "/home/user1/.conda/envs/memgpt/lib/python3.11/site-packages/autogen/agentchat/conversable_agent.py", line 117, in init
self.client = OpenAIWrapper(**self.llm_config)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/home/user1/.conda/envs/memgpt/lib/python3.11/site-packages/autogen/oai/client.py", line 83, in init
self._clients = [self._client(extra_kwargs, openai_config)]
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/home/user1/.conda/envs/memgpt/lib/python3.11/site-packages/autogen/oai/client.py", line 138, in _client
client = OpenAI(**openai_config)
^^^^^^^^^^^^^^^^^^^^^^^
File "/home/user1/.conda/envs/memgpt/lib/python3.11/site-packages/openai/_client.py", line 92, in init
raise OpenAIError(
openai.OpenAIError: The api_key client option must be set either by passing api_key to the client or by setting the OPENAI_API_KEY environment variable

Please describe your setup
OS: Ubuntu 22.04 desktop
HW: HP Z8 Fury with 4 RTX 6000 ADAs and 1TB RAM
python version = 3.11.3 (I have tried the 3.10.9) with Conda

• MemGPT version
  • What is the output of memgpt version? 0.2.10
• How did you install memgpt?
  • pip install pymemgpt and from source
• Describe your setup
  • What's your OS (Windows/MacOS/Linux)? Linux (Ubuntu 22.04)
  • How are you running memgpt? (cmd.exe/Powershell/Anaconda Shell/Terminal) Terminal and also SSH from a windows machine

Screenshots
If applicable, add screenshots to help explain your problem.

Additional context
Add any other context about the problem here.

---

If you're not using OpenAI, please provide additional information on your local LLM setup:

Local LLM details

If you are trying to run MemGPT with local LLMs, please provide the following information:

• The exact model you're trying to use (e.g. dolphin-2.1-mistral-7b.Q6_K.gguf)
• The local LLM backend you are using (web UI? LM Studio?) tried both
• Your hardware for the local LLM backend (local computer? operating system? remote RunPod?) Local computer (Z8 Fury)

[naseerfaheem]

I tried creating a new environment with all the recommended steps, still the same api error. Here is the full error:

OpenAIError Traceback (most recent call last)
Cell In[13], line 11
3 coder = autogen.AssistantAgent(
4 name="Coder",
5 llm_config=llm_config,
6 )
8 else:
9 # In our example, we swap this AutoGen agent with a MemGPT agent
10 # This MemGPT agent will have all the benefits of MemGPT, ie persistent memory, etc.
---> 11 coder = create_memgpt_autogen_agent_from_config(
12 "MemGPT_coder",
13 llm_config=llm_config_memgpt,
14 system_message=f"I am a 10x engineer, trained in Python. I was the first engineer at Uber "
15 f"(which I make sure to tell everyone I work with).\n"
16 f"You are participating in a group chat with a user ({user_proxy.name}) "
17 f"and a product manager ({pm.name}).",
18 interface_kwargs=interface_kwargs,
19 default_auto_reply="...", # Set a default auto-reply message here (non-empty auto-reply is required for LM Studio)
20 skip_verify=False, # NOTE: you should set this to True if you expect your MemGPT AutoGen agent to call a function other than send_message on the first turn
21 )
23 # Initialize the group chat between the user and two LLM agents (PM and coder)
24 groupchat = autogen.GroupChat(agents=[user_proxy, pm, coder], messages=[], max_round=12)

File ~/Documents/DataScience/installs/MemGPT/memgpt/autogen/memgpt_agent.py:95, in create_memgpt_autogen_agent_from_config(name, system_message, is_termination_msg, max_consecutive_auto_reply, human_input_mode, function_map, code_execution_config, llm_config, nonmemgpt_llm_config, default_auto_reply, interface_kwargs, skip_verify)
92 if function_map is not None or code_execution_config is not None:
93 raise NotImplementedError
---> 95 autogen_memgpt_agent = create_autogen_memgpt_agent(
96 agent_config,
97 default_auto_reply=default_auto_reply,
98 is_termination_msg=is_termination_msg,
99 interface_kwargs=interface_kwargs,
100 skip_verify=skip_verify,
101 )
103 if human_input_mode != "ALWAYS":
104 coop_agent1 = create_autogen_memgpt_agent(
105 agent_config,
106 default_auto_reply=default_auto_reply,
(...)
109 skip_verify=skip_verify,
110 )

File ~/Documents/DataScience/installs/MemGPT/memgpt/autogen/memgpt_agent.py:184, in create_autogen_memgpt_agent(agent_config, skip_verify, interface, interface_kwargs, persistence_manager, persistence_manager_kwargs, default_auto_reply, is_termination_msg)
172 persistence_manager = LocalStateManager(**persistence_manager_kwargs) if persistence_manager is None else persistence_manager
174 memgpt_agent = presets.use_preset(
175 agent_config.preset,
176 agent_config,
(...)
181 persistence_manager,
182 )
--> 184 autogen_memgpt_agent = MemGPTAgent(
185 name=agent_config.name,
186 agent=memgpt_agent,
187 default_auto_reply=default_auto_reply,
188 is_termination_msg=is_termination_msg,
189 skip_verify=skip_verify,
190 )
191 return autogen_memgpt_agent

File ~/Documents/DataScience/installs/MemGPT/memgpt/autogen/memgpt_agent.py:204, in MemGPTAgent.init(self, name, agent, skip_verify, concat_other_agent_messages, is_termination_msg, default_auto_reply)
195 def init(
196 self,
197 name: str,
(...)
202 default_auto_reply: Optional[Union[str, Dict, None]] = "",
203 ):
--> 204 super().init(name)
205 self.agent = agent
206 self.skip_verify = skip_verify

File ~/.conda/envs/memgpt/lib/python3.11/site-packages/autogen/agentchat/conversable_agent.py:117, in ConversableAgent.init(self, name, system_message, is_termination_msg, max_consecutive_auto_reply, human_input_mode, function_map, code_execution_config, llm_config, default_auto_reply)
115 if isinstance(llm_config, dict):
116 self.llm_config.update(llm_config)
--> 117 self.client = OpenAIWrapper(**self.llm_config)
119 self._code_execution_config: Union[Dict, Literal[False]] = (
120 {} if code_execution_config is None else code_execution_config
121 )
122 self.human_input_mode = human_input_mode

File ~/.conda/envs/memgpt/lib/python3.11/site-packages/autogen/oai/client.py:83, in OpenAIWrapper.init(self, config_list, **base_config)
78 self._config_list = [
79 {**extra_kwargs, **{k: v for k, v in config.items() if k not in self.openai_kwargs}}
80 for config in config_list
81 ]
82 else:
---> 83 self._clients = [self._client(extra_kwargs, openai_config)]
84 self._config_list = [extra_kwargs]

File ~/.conda/envs/memgpt/lib/python3.11/site-packages/autogen/oai/client.py:138, in OpenAIWrapper._client(self, config, openai_config)
136 openai_config = {**openai_config, **{k: v for k, v in config.items() if k in self.openai_kwargs}}
137 self._process_for_azure(openai_config, config)
--> 138 client = OpenAI(**openai_config)
139 return client

File ~/.conda/envs/memgpt/lib/python3.11/site-packages/openai/_client.py:92, in OpenAI.init(self, api_key, organization, base_url, timeout, max_retries, default_headers, default_query, http_client, _strict_response_validation)
90 api_key = os.environ.get("OPENAI_API_KEY")
91 if api_key is None:
---> 92 raise OpenAIError(
93 "The api_key client option must be set either by passing api_key to the client or by setting the OPENAI_API_KEY environment variable"
...
94 )
95 self.api_key = api_key
97 if organization is None:

OpenAIError: The api_key client option must be set either by passing api_key to the client or by setting the OPENAI_API_KEY environment variable

[cpacker]

This error is happening because AutoGen uses the openai package (even when using local LLMs), and the openai package needs OPENAI_API_KEY to be set to something, even if it's a dummy variable.

First do:

export OPENAI_API_KEY="null"

Then do:

python agent_docs.py