search

OpenBMB / ChatDev

Create Customized Software using Natural Language Idea (through LLM-powered Multi-Agent Collaboration)
https://arxiv.org/abs/2307.07924
Apache License 2.0
25.67k stars 3.23k forks source link

Integrate ChatDev with Langchain #401

Open M340i opened 4 months ago

M340i commented 4 months ago

I noticed that since there is no memory hosting the past work done by the custom multi-agent, is it possible to merge the Langchain to store the input/output of past records for each agent to use in future output generation?

Linuxscot1974 commented 4 months ago

not without file injection @run.py time which would hold a tokenized txt file of the conversation history. It would become part of the task_prompt file along with a whole prompt engineering construct, much like the ones we use today except more in depth and increased ability to create rules for an agent. this is not in the package as of yet. I have started my own run.py version with a link to the file in first draft. Issue #321 In the end it will inject files where we have to type task, name, org.

M340i commented 4 months ago

So will it requires more token during the task creation?

Linuxscot1974 commented 4 months ago

that would depend on the specificity of the task outline. more information will define the task better.

M340i commented 4 months ago

@Linuxscot1974 may I know have you tried out the Redis chatGPT memory and if yes, what are the differences between langchain and redis chatGPT memory?

M340i commented 4 months ago

In the end it will inject files where we have to type task, name, org.

So it means that we ended up having more parameters for output creation?