Persisting Chat History with the React Agent

Question

how do i persist chat history when using the react agent? the docs doesn't mention anything like that. I need the agent to still remember its previous outputs

Logan M · Answer

You can eitherpersist it yourself (agent.chat_history to get the chat history, you can pass it in with agent.chat("hello", chat_history=chat_history). Its a list of ChatMessage objects, which are pydantic, so serializableor you can use a chat store https://docs.llamaindex.ai/en/stable/module_guides/storing/chat_stores/#chat-stores

4bidden · Answer

is chat_history an import? or its just the agent.chat_history variable @Logan M

Logan M · Answer

just an attribute on the agent

Logan M · Answer

it returns the current list of chat messages in the memory

Logan M · Answer

try it out 🙂

4bidden · Answer

not working, keeps running the same function. generate_wallet_tool = FunctionTool.from_defaults(fn=generate_solana_wallet) # Wrap the transaction execution function
# execute_transaction_tool = FunctionTool.from_defaults(fn=execute_solana_transaction) # Create the agent with the tools, memory, and LLM
agent = ReActAgent.from_tools([generate_wallet_tool], verbose=True)
agent.chat_history
response = agent.chat("what was the previous wallet address?")

Logan M · Answer

Not sure I understand the code you are running. You didn't use the chat history anywhere?

In the same script, it will maintain memory. But if the program/script exits, you need to store it somewhere and restore it

4bidden · Answer

oh

4bidden · Answer

meaning a chatstore is needed

Logan M · Answer

or you an just dump it whereever you want (since the chat messages are json serializable)

Logan M · Answer

chat_history = agent.chat_history
chat_history_dicts = [x.model_dump() for x in chat_history] # dump in redis, mongodb, json.dump to disk, pickle, etc... # reload it
from llama_index.core.llms import ChatMessage chat_history = [ChatMessage.model_validate(x) for x in chat_history_dicts]
agent.chat("Hello", chat_history=chat_history)

Logan M · Answer

Or, you can use the chat store approach yes

4bidden · Answer

Thanks

Find answers from the community

Persisting Chat History with the React Agent