langchain/cookbook/sharedmemory_for_tools.ipynb

Shared memory across agents and tools

This notebook goes over adding memory to both an Agent and its tools. Before going through this notebook, please walk through the following notebooks, as this will build on top of both of them:

• Adding memory to an LLM Chain
• Custom Agents

We are going to create a custom Agent. The agent has access to a conversation memory, search tool, and a summarization tool. The summarization tool also needs access to the conversation memory.

In [1]:

from langchain import hub
from langchain.agents import AgentExecutor, Tool, ZeroShotAgent, create_react_agent
from langchain.chains import LLMChain
from langchain.memory import ConversationBufferMemory, ReadOnlySharedMemory
from langchain.prompts import PromptTemplate
from langchain_community.utilities import GoogleSearchAPIWrapper
from langchain_openai import OpenAI

In [2]:

template = """This is a conversation between a human and a bot:

{chat_history}

Write a summary of the conversation for {input}:
"""

prompt = PromptTemplate(input_variables=["input", "chat_history"], template=template)
memory = ConversationBufferMemory(memory_key="chat_history")
readonlymemory = ReadOnlySharedMemory(memory=memory)
summary_chain = LLMChain(
    llm=OpenAI(),
    prompt=prompt,
    verbose=True,
    memory=readonlymemory,  # use the read-only memory to prevent the tool from modifying the memory
)

In [3]:

search = GoogleSearchAPIWrapper()
tools = [
    Tool(
        name="Search",
        func=search.run,
        description="useful for when you need to answer questions about current events",
    ),
    Tool(
        name="Summary",
        func=summary_chain.run,
        description="useful for when you summarize a conversation. The input to this tool should be a string, representing who will read this summary.",
    ),
]

In [4]:

prompt = hub.pull("hwchase17/react")

We can now construct the LLMChain, with the Memory object, and then create the agent.

In [5]:

model = OpenAI()
agent = create_react_agent(model, tools, prompt)
agent_executor = AgentExecutor(agent=agent, tools=tools, memory=memory)

In [36]:

agent_executor.invoke({"input": "What is ChatGPT?"})

Out [36]:

> Entering new AgentExecutor chain...
Thought: I should research ChatGPT to answer this question.
Action: Search
Action Input: "ChatGPT"
Observation: Nov 30, 2022 ... We've trained a model called ChatGPT which interacts in a conversational way. The dialogue format makes it possible for ChatGPT to answer ... ChatGPT is an artificial intelligence chatbot developed by OpenAI and launched in November 2022. It is built on top of OpenAI's GPT-3 family of large ... ChatGPT. We've trained a model called ChatGPT which interacts in a conversational way. The dialogue format makes it possible for ChatGPT to answer ... Feb 2, 2023 ... ChatGPT, the popular chatbot from OpenAI, is estimated to have reached 100 million monthly active users in January, just two months after ... 2 days ago ... ChatGPT recently launched a new version of its own plagiarism detection tool, with hopes that it will squelch some of the criticism around how ... An API for accessing new AI models developed by OpenAI. Feb 19, 2023 ... ChatGPT is an AI chatbot system that OpenAI released in November to show off and test what a very large, powerful AI system can accomplish. You ... ChatGPT is fine-tuned from GPT-3.5, a language model trained to produce text. ChatGPT was optimized for dialogue by using Reinforcement Learning with Human ... 3 days ago ... Visual ChatGPT connects ChatGPT and a series of Visual Foundation Models to enable sending and receiving images during chatting. Dec 1, 2022 ... ChatGPT is a natural language processing tool driven by AI technology that allows you to have human-like conversations and much more with a ...
Thought: I now know the final answer.
Final Answer: ChatGPT is an artificial intelligence chatbot developed by OpenAI and launched in November 2022. It is built on top of OpenAI's GPT-3 family of large language models and is optimized for dialogue by using Reinforcement Learning with Human-in-the-Loop. It is also capable of sending and receiving images during chatting.

> Finished chain.

"ChatGPT is an artificial intelligence chatbot developed by OpenAI and launched in November 2022. It is built on top of OpenAI's GPT-3 family of large language models and is optimized for dialogue by using Reinforcement Learning with Human-in-the-Loop. It is also capable of sending and receiving images during chatting."

---------------------------------------------------------------------------
KeyboardInterrupt                         Traceback (most recent call last)
Cell In[36], line 1
----> 1 agent_executor.invoke({"input":"What is ChatGPT?"})

File ~/code/langchain/libs/langchain/langchain/chains/base.py:163, in Chain.invoke(self, input, config, **kwargs)
    161 except BaseException as e:
    162     run_manager.on_chain_error(e)
--> 163     raise e
    164 run_manager.on_chain_end(outputs)
    166 if include_run_info:

File ~/code/langchain/libs/langchain/langchain/chains/base.py:153, in Chain.invoke(self, input, config, **kwargs)
    150 try:
    151     self._validate_inputs(inputs)
    152     outputs = (
--> 153         self._call(inputs, run_manager=run_manager)
    154         if new_arg_supported
    155         else self._call(inputs)
    156     )
    158     final_outputs: Dict[str, Any] = self.prep_outputs(
    159         inputs, outputs, return_only_outputs
    160     )
    161 except BaseException as e:

File ~/code/langchain/libs/langchain/langchain/agents/agent.py:1432, in AgentExecutor._call(self, inputs, run_manager)
   1430 # We now enter the agent loop (until it returns something).
   1431 while self._should_continue(iterations, time_elapsed):
-> 1432     next_step_output = self._take_next_step(
   1433         name_to_tool_map,
   1434         color_mapping,
   1435         inputs,
   1436         intermediate_steps,
   1437         run_manager=run_manager,
   1438     )
   1439     if isinstance(next_step_output, AgentFinish):
   1440         return self._return(
   1441             next_step_output, intermediate_steps, run_manager=run_manager
   1442         )

File ~/code/langchain/libs/langchain/langchain/agents/agent.py:1138, in AgentExecutor._take_next_step(self, name_to_tool_map, color_mapping, inputs, intermediate_steps, run_manager)
   1129 def _take_next_step(
   1130     self,
   1131     name_to_tool_map: Dict[str, BaseTool],
   (...)
   1135     run_manager: Optional[CallbackManagerForChainRun] = None,
   1136 ) -> Union[AgentFinish, List[Tuple[AgentAction, str]]]:
   1137     return self._consume_next_step(
-> 1138         [
   1139             a
   1140             for a in self._iter_next_step(
   1141                 name_to_tool_map,
   1142                 color_mapping,
   1143                 inputs,
   1144                 intermediate_steps,
   1145                 run_manager,
   1146             )
   1147         ]
   1148     )

File ~/code/langchain/libs/langchain/langchain/agents/agent.py:1138, in <listcomp>(.0)
   1129 def _take_next_step(
   1130     self,
   1131     name_to_tool_map: Dict[str, BaseTool],
   (...)
   1135     run_manager: Optional[CallbackManagerForChainRun] = None,
   1136 ) -> Union[AgentFinish, List[Tuple[AgentAction, str]]]:
   1137     return self._consume_next_step(
-> 1138         [
   1139             a
   1140             for a in self._iter_next_step(
   1141                 name_to_tool_map,
   1142                 color_mapping,
   1143                 inputs,
   1144                 intermediate_steps,
   1145                 run_manager,
   1146             )
   1147         ]
   1148     )

File ~/code/langchain/libs/langchain/langchain/agents/agent.py:1223, in AgentExecutor._iter_next_step(self, name_to_tool_map, color_mapping, inputs, intermediate_steps, run_manager)
   1221     yield agent_action
   1222 for agent_action in actions:
-> 1223     yield self._perform_agent_action(
   1224         name_to_tool_map, color_mapping, agent_action, run_manager
   1225     )

File ~/code/langchain/libs/langchain/langchain/agents/agent.py:1245, in AgentExecutor._perform_agent_action(self, name_to_tool_map, color_mapping, agent_action, run_manager)
   1243         tool_run_kwargs["llm_prefix"] = ""
   1244     # We then call the tool on the tool input to get an observation
-> 1245     observation = tool.run(
   1246         agent_action.tool_input,
   1247         verbose=self.verbose,
   1248         color=color,
   1249         callbacks=run_manager.get_child() if run_manager else None,
   1250         **tool_run_kwargs,
   1251     )
   1252 else:
   1253     tool_run_kwargs = self.agent.tool_run_logging_kwargs()

File ~/code/langchain/libs/core/langchain_core/tools.py:422, in BaseTool.run(self, tool_input, verbose, start_color, color, callbacks, tags, metadata, run_name, run_id, **kwargs)
    420 except (Exception, KeyboardInterrupt) as e:
    421     run_manager.on_tool_error(e)
--> 422     raise e
    423 else:
    424     run_manager.on_tool_end(observation, color=color, name=self.name, **kwargs)

File ~/code/langchain/libs/core/langchain_core/tools.py:381, in BaseTool.run(self, tool_input, verbose, start_color, color, callbacks, tags, metadata, run_name, run_id, **kwargs)
    378     parsed_input = self._parse_input(tool_input)
    379     tool_args, tool_kwargs = self._to_args_and_kwargs(parsed_input)
    380     observation = (
--> 381         self._run(*tool_args, run_manager=run_manager, **tool_kwargs)
    382         if new_arg_supported
    383         else self._run(*tool_args, **tool_kwargs)
    384     )
    385 except ValidationError as e:
    386     if not self.handle_validation_error:

File ~/code/langchain/libs/core/langchain_core/tools.py:588, in Tool._run(self, run_manager, *args, **kwargs)
    579 if self.func:
    580     new_argument_supported = signature(self.func).parameters.get("callbacks")
    581     return (
    582         self.func(
    583             *args,
    584             callbacks=run_manager.get_child() if run_manager else None,
    585             **kwargs,
    586         )
    587         if new_argument_supported
--> 588         else self.func(*args, **kwargs)
    589     )
    590 raise NotImplementedError("Tool does not support sync")

File ~/code/langchain/libs/community/langchain_community/utilities/google_search.py:94, in GoogleSearchAPIWrapper.run(self, query)
     92 """Run query through GoogleSearch and parse result."""
     93 snippets = []
---> 94 results = self._google_search_results(query, num=self.k)
     95 if len(results) == 0:
     96     return "No good Google Search Result was found"

File ~/code/langchain/libs/community/langchain_community/utilities/google_search.py:62, in GoogleSearchAPIWrapper._google_search_results(self, search_term, **kwargs)
     60 if self.siterestrict:
     61     cse = cse.siterestrict()
---> 62 res = cse.list(q=search_term, cx=self.google_cse_id, **kwargs).execute()
     63 return res.get("items", [])

File ~/code/langchain/.venv/lib/python3.10/site-packages/googleapiclient/_helpers.py:130, in positional.<locals>.positional_decorator.<locals>.positional_wrapper(*args, **kwargs)
    128     elif positional_parameters_enforcement == POSITIONAL_WARNING:
    129         logger.warning(message)
--> 130 return wrapped(*args, **kwargs)

File ~/code/langchain/.venv/lib/python3.10/site-packages/googleapiclient/http.py:923, in HttpRequest.execute(self, http, num_retries)
    920     self.headers["content-length"] = str(len(self.body))
    922 # Handle retries for server-side errors.
--> 923 resp, content = _retry_request(
    924     http,
    925     num_retries,
    926     "request",
    927     self._sleep,
    928     self._rand,
    929     str(self.uri),
    930     method=str(self.method),
    931     body=self.body,
    932     headers=self.headers,
    933 )
    935 for callback in self.response_callbacks:
    936     callback(resp)

File ~/code/langchain/.venv/lib/python3.10/site-packages/googleapiclient/http.py:191, in _retry_request(http, num_retries, req_type, sleep, rand, uri, method, *args, **kwargs)
    189 try:
    190     exception = None
--> 191     resp, content = http.request(uri, method, *args, **kwargs)
    192 # Retry on SSL errors and socket timeout errors.
    193 except _ssl_SSLError as ssl_error:

File ~/code/langchain/.venv/lib/python3.10/site-packages/httplib2/__init__.py:1724, in Http.request(self, uri, method, body, headers, redirections, connection_type)
   1722             content = b""
   1723         else:
-> 1724             (response, content) = self._request(
   1725                 conn, authority, uri, request_uri, method, body, headers, redirections, cachekey,
   1726             )
   1727 except Exception as e:
   1728     is_timeout = isinstance(e, socket.timeout)

File ~/code/langchain/.venv/lib/python3.10/site-packages/httplib2/__init__.py:1444, in Http._request(self, conn, host, absolute_uri, request_uri, method, body, headers, redirections, cachekey)
   1441 if auth:
   1442     auth.request(method, request_uri, headers, body)
-> 1444 (response, content) = self._conn_request(conn, request_uri, method, body, headers)
   1446 if auth:
   1447     if auth.response(response, body):

File ~/code/langchain/.venv/lib/python3.10/site-packages/httplib2/__init__.py:1366, in Http._conn_request(self, conn, request_uri, method, body, headers)
   1364 try:
   1365     if conn.sock is None:
-> 1366         conn.connect()
   1367     conn.request(method, request_uri, body, headers)
   1368 except socket.timeout:

File ~/code/langchain/.venv/lib/python3.10/site-packages/httplib2/__init__.py:1156, in HTTPSConnectionWithTimeout.connect(self)
   1154 if has_timeout(self.timeout):
   1155     sock.settimeout(self.timeout)
-> 1156 sock.connect((self.host, self.port))
   1158 self.sock = self._context.wrap_socket(sock, server_hostname=self.host)
   1160 # Python 3.3 compatibility: emulate the check_hostname behavior

KeyboardInterrupt: 

To test the memory of this agent, we can ask a followup question that relies on information in the previous exchange to be answered correctly.

In [7]:

agent_executor.invoke({"input": "Who developed it?"})

Out [7]:

> Entering new AgentExecutor chain...
Thought: I need to find out who developed ChatGPT
Action: Search
Action Input: Who developed ChatGPT
Observation: ChatGPT is an artificial intelligence chatbot developed by OpenAI and launched in November 2022. It is built on top of OpenAI's GPT-3 family of large ... Feb 15, 2023 ... Who owns Chat GPT? Chat GPT is owned and developed by AI research and deployment company, OpenAI. The organization is headquartered in San ... Feb 8, 2023 ... ChatGPT is an AI chatbot developed by San Francisco-based startup OpenAI. OpenAI was co-founded in 2015 by Elon Musk and Sam Altman and is ... Dec 7, 2022 ... ChatGPT is an AI chatbot designed and developed by OpenAI. The bot works by generating text responses based on human-user input, like questions ... Jan 12, 2023 ... In 2019, Microsoft invested $1 billion in OpenAI, the tiny San Francisco company that designed ChatGPT. And in the years since, it has quietly ... Jan 25, 2023 ... The inside story of ChatGPT: How OpenAI founder Sam Altman built the world's hottest technology with billions from Microsoft. Dec 3, 2022 ... ChatGPT went viral on social media for its ability to do anything from code to write essays. · The company that created the AI chatbot has a ... Jan 17, 2023 ... While many Americans were nursing hangovers on New Year's Day, 22-year-old Edward Tian was working feverishly on a new app to combat misuse ... ChatGPT is a language model created by OpenAI, an artificial intelligence research laboratory consisting of a team of researchers and engineers focused on ... 1 day ago ... Everyone is talking about ChatGPT, developed by OpenAI. This is such a great tool that has helped to make AI more accessible to a wider ...
Thought: I now know the final answer
Final Answer: ChatGPT was developed by OpenAI.

> Finished chain.

'ChatGPT was developed by OpenAI.'

In [8]:

agent_executor.invoke(
    {"input": "Thanks. Summarize the conversation, for my daughter 5 years old."}
)

Out [8]:

> Entering new AgentExecutor chain...
Thought: I need to simplify the conversation for a 5 year old.
Action: Summary
Action Input: My daughter 5 years old

> Entering new LLMChain chain...
Prompt after formatting:
This is a conversation between a human and a bot:

Human: What is ChatGPT?
AI: ChatGPT is an artificial intelligence chatbot developed by OpenAI and launched in November 2022. It is built on top of OpenAI's GPT-3 family of large language models and is optimized for dialogue by using Reinforcement Learning with Human-in-the-Loop. It is also capable of sending and receiving images during chatting.
Human: Who developed it?
AI: ChatGPT was developed by OpenAI.

Write a summary of the conversation for My daughter 5 years old:


> Finished chain.

Observation: 
The conversation was about ChatGPT, an artificial intelligence chatbot. It was created by OpenAI and can send and receive images while chatting.
Thought: I now know the final answer.
Final Answer: ChatGPT is an artificial intelligence chatbot created by OpenAI that can send and receive images while chatting.

> Finished chain.

'ChatGPT is an artificial intelligence chatbot created by OpenAI that can send and receive images while chatting.'

Confirm that the memory was correctly updated.

In [9]:

print(agent_executor.memory.buffer)

Human: What is ChatGPT?
AI: ChatGPT is an artificial intelligence chatbot developed by OpenAI and launched in November 2022. It is built on top of OpenAI's GPT-3 family of large language models and is optimized for dialogue by using Reinforcement Learning with Human-in-the-Loop. It is also capable of sending and receiving images during chatting.
Human: Who developed it?
AI: ChatGPT was developed by OpenAI.
Human: Thanks. Summarize the conversation, for my daughter 5 years old.
AI: ChatGPT is an artificial intelligence chatbot created by OpenAI that can send and receive images while chatting.

For comparison, below is a bad example that uses the same memory for both the Agent and the tool.

In [10]:

## This is a bad practice for using the memory.
## Use the ReadOnlySharedMemory class, as shown above.

template = """This is a conversation between a human and a bot:

{chat_history}

Write a summary of the conversation for {input}:
"""

prompt = PromptTemplate(input_variables=["input", "chat_history"], template=template)
memory = ConversationBufferMemory(memory_key="chat_history")
summary_chain = LLMChain(
    llm=OpenAI(),
    prompt=prompt,
    verbose=True,
    memory=memory,  # <--- this is the only change
)

search = GoogleSearchAPIWrapper()
tools = [
    Tool(
        name="Search",
        func=search.run,
        description="useful for when you need to answer questions about current events",
    ),
    Tool(
        name="Summary",
        func=summary_chain.run,
        description="useful for when you summarize a conversation. The input to this tool should be a string, representing who will read this summary.",
    ),
]

prompt = hub.pull("hwchase17/react")
agent = create_react_agent(model, tools, prompt)
agent_executor = AgentExecutor(agent=agent, tools=tools, memory=memory)

In [11]:

agent_executor.invoke({"input": "What is ChatGPT?"})

Out [11]:

> Entering new AgentExecutor chain...
Thought: I should research ChatGPT to answer this question.
Action: Search
Action Input: "ChatGPT"
Observation: Nov 30, 2022 ... We've trained a model called ChatGPT which interacts in a conversational way. The dialogue format makes it possible for ChatGPT to answer ... ChatGPT is an artificial intelligence chatbot developed by OpenAI and launched in November 2022. It is built on top of OpenAI's GPT-3 family of large ... ChatGPT. We've trained a model called ChatGPT which interacts in a conversational way. The dialogue format makes it possible for ChatGPT to answer ... Feb 2, 2023 ... ChatGPT, the popular chatbot from OpenAI, is estimated to have reached 100 million monthly active users in January, just two months after ... 2 days ago ... ChatGPT recently launched a new version of its own plagiarism detection tool, with hopes that it will squelch some of the criticism around how ... An API for accessing new AI models developed by OpenAI. Feb 19, 2023 ... ChatGPT is an AI chatbot system that OpenAI released in November to show off and test what a very large, powerful AI system can accomplish. You ... ChatGPT is fine-tuned from GPT-3.5, a language model trained to produce text. ChatGPT was optimized for dialogue by using Reinforcement Learning with Human ... 3 days ago ... Visual ChatGPT connects ChatGPT and a series of Visual Foundation Models to enable sending and receiving images during chatting. Dec 1, 2022 ... ChatGPT is a natural language processing tool driven by AI technology that allows you to have human-like conversations and much more with a ...
Thought: I now know the final answer.
Final Answer: ChatGPT is an artificial intelligence chatbot developed by OpenAI and launched in November 2022. It is built on top of OpenAI's GPT-3 family of large language models and is optimized for dialogue by using Reinforcement Learning with Human-in-the-Loop. It is also capable of sending and receiving images during chatting.

> Finished chain.

"ChatGPT is an artificial intelligence chatbot developed by OpenAI and launched in November 2022. It is built on top of OpenAI's GPT-3 family of large language models and is optimized for dialogue by using Reinforcement Learning with Human-in-the-Loop. It is also capable of sending and receiving images during chatting."

In [12]:

agent_executor.invoke({"input": "Who developed it?"})

Out [12]:

> Entering new AgentExecutor chain...
Thought: I need to find out who developed ChatGPT
Action: Search
Action Input: Who developed ChatGPT
Observation: ChatGPT is an artificial intelligence chatbot developed by OpenAI and launched in November 2022. It is built on top of OpenAI's GPT-3 family of large ... Feb 15, 2023 ... Who owns Chat GPT? Chat GPT is owned and developed by AI research and deployment company, OpenAI. The organization is headquartered in San ... Feb 8, 2023 ... ChatGPT is an AI chatbot developed by San Francisco-based startup OpenAI. OpenAI was co-founded in 2015 by Elon Musk and Sam Altman and is ... Dec 7, 2022 ... ChatGPT is an AI chatbot designed and developed by OpenAI. The bot works by generating text responses based on human-user input, like questions ... Jan 12, 2023 ... In 2019, Microsoft invested $1 billion in OpenAI, the tiny San Francisco company that designed ChatGPT. And in the years since, it has quietly ... Jan 25, 2023 ... The inside story of ChatGPT: How OpenAI founder Sam Altman built the world's hottest technology with billions from Microsoft. Dec 3, 2022 ... ChatGPT went viral on social media for its ability to do anything from code to write essays. · The company that created the AI chatbot has a ... Jan 17, 2023 ... While many Americans were nursing hangovers on New Year's Day, 22-year-old Edward Tian was working feverishly on a new app to combat misuse ... ChatGPT is a language model created by OpenAI, an artificial intelligence research laboratory consisting of a team of researchers and engineers focused on ... 1 day ago ... Everyone is talking about ChatGPT, developed by OpenAI. This is such a great tool that has helped to make AI more accessible to a wider ...
Thought: I now know the final answer
Final Answer: ChatGPT was developed by OpenAI.

> Finished chain.

'ChatGPT was developed by OpenAI.'

In [13]:

agent_executor.invoke(
    {"input": "Thanks. Summarize the conversation, for my daughter 5 years old."}
)

Out [13]:

> Entering new AgentExecutor chain...
Thought: I need to simplify the conversation for a 5 year old.
Action: Summary
Action Input: My daughter 5 years old

> Entering new LLMChain chain...
Prompt after formatting:
This is a conversation between a human and a bot:

Human: What is ChatGPT?
AI: ChatGPT is an artificial intelligence chatbot developed by OpenAI and launched in November 2022. It is built on top of OpenAI's GPT-3 family of large language models and is optimized for dialogue by using Reinforcement Learning with Human-in-the-Loop. It is also capable of sending and receiving images during chatting.
Human: Who developed it?
AI: ChatGPT was developed by OpenAI.

Write a summary of the conversation for My daughter 5 years old:


> Finished chain.

Observation: 
The conversation was about ChatGPT, an artificial intelligence chatbot developed by OpenAI. It is designed to have conversations with humans and can also send and receive images.
Thought: I now know the final answer.
Final Answer: ChatGPT is an artificial intelligence chatbot developed by OpenAI that can have conversations with humans and send and receive images.

> Finished chain.

'ChatGPT is an artificial intelligence chatbot developed by OpenAI that can have conversations with humans and send and receive images.'

The final answer is not wrong, but we see the 3rd Human input is actually from the agent in the memory because the memory was modified by the summary tool.

In [14]:

print(agent_executor.memory.buffer)

Human: What is ChatGPT?
AI: ChatGPT is an artificial intelligence chatbot developed by OpenAI and launched in November 2022. It is built on top of OpenAI's GPT-3 family of large language models and is optimized for dialogue by using Reinforcement Learning with Human-in-the-Loop. It is also capable of sending and receiving images during chatting.
Human: Who developed it?
AI: ChatGPT was developed by OpenAI.
Human: My daughter 5 years old
AI: 
The conversation was about ChatGPT, an artificial intelligence chatbot developed by OpenAI. It is designed to have conversations with humans and can also send and receive images.
Human: Thanks. Summarize the conversation, for my daughter 5 years old.
AI: ChatGPT is an artificial intelligence chatbot developed by OpenAI that can have conversations with humans and send and receive images.