Given that ChatGPT was trained using data only up to September 2021 and has a context limit of 4K tokens per call to the LLM, we encounter the following constraints:
- It's not suitable for discussions based on content produced after this date, such as understanding a newly developed API or our company's updated documentation.
- Directly pasting extensive content is restricted due to the context limitation.
Place your files of any format into the designated folder. Once there, the data will be processed and integrated into a vector database. You can then interact with the chatbot using the Streamlit web application framework. Notably, the chatbot is also equipped to answer queries based on data that can be sourced from a Google search.
The chatbot's capabilities are backed by an agent that utilizes several tools:
-
Company DB: Enables question-answering derived from the company's database embeddings.
-
Boto3 DB: Designed for question-answering using the provided Boto3 documentation
The following methods are supported -
- using documents loader - processes the provided documents
- using scraping loader - scrapes all the relevant html pages of a specific boto3 service and loads them to a vector DB
-
Google Search Tool: Useful for fetching information from the web.
Clone the repository, set up the virtual environment, and install the required packages
-
git clone git@github.com:nirbar1985/compibot.git
-
( In case you have python version 3.11.4 installed in pyenv)
pyenv local 3.11.4 -
Install dependencies
poetry install
-
Enter virtual env by:
poetry shell
- Create .env file
- Place your OPENAI_API_KEY into .env file
- Place your SERPAPI_API_KEY into .env file
- The format should be -
OPENAI_API_KEY= SERPAPI_API_KEY=
-
Place the company documentation files into the designated directory inside documents directory
-
Using Boto3 Documents loader - place the boto3 documentation files into the docs_boto3 directory inside documents directory
-
run once the indexing script:
python indexing.py --process-boto3-docs --boto3-loader documents_loader -
Using Boto3 Scraping loader - run once the indexing script using this format:
python indexing.py --process-boto3-docs --boto3-loader scraping_loader --boto3-service-name rolesanywhereAfter runing this script, all the relevant html pages regarding the APIs of a specific boto3 service (in this example rolesanywhere service) were scraped and loaded into a vector DB
-
The vector database is now stored persistently in the "db" directory.
Kick of the chatbot by running:
streamlit run chatbot_ui.py
Because a known bug in langchain output parser in some scenarios - sometimes when the output parser gets text with code snippet, you might get in the chatbot this exception - raise OutputParserException(f"Could not parse LLM output: {text}") from e
please paste the following temporarily fix:
except Exception as e:
# If any other exception is raised during parsing, also raise an
# OutputParserException
try:
response = json.loads(text)
return AgentFinish({"output": response["action_input"]}, text)
except json.JSONDecodeError:
# Handle JSON parsing error, if needed
raise OutputParserException(f"Could not parse LLM output: {text}") from e
in this location - https://github.com/langchain-ai/langchain/blob/master/libs/langchain/langchain/agents/conversational_chat/output_parser.py#L50
Contributions are what make the open source community such an amazing place to learn, inspire, and create. Any contributions you make are greatly appreciated.
If you have a suggestion that would make this better, please fork the repo and create a pull request. You can also simply open an issue with the tag "enhancement". Don't forget to give the project a star! Thanks again!
Fork the Project Create your Feature Branch (git checkout -b feature/AmazingFeature) Commit your Changes (git commit -m 'Add some AmazingFeature') Push to the Branch (git push origin feature/AmazingFeature) Open a Pull Request (back to top)
Distributed under the MIT License. See LICENSE.txt for more information.