[Question]: chatbot to facilitate development.

[buckmaxwell]

Question Validation

• I have searched both the documentation and discord for an answer.

Question

What I want to do is simple - at the root of a code repository run the following.

My expectation would have been that I could then ask questions about the code and it would be kind of like chat gpt 4 but I wouldn't need to provide all the extra context.

It isn't working that way though - it doesn't seem to be able to answer any questions - as if the indexer was only built for understanding text files and docs not for reading code. Is there another reader or indexer I could be using? Or another way to solve this problem?

import os

os.environ["OPENAI_API_KEY"] = os.environ.get("OPENAI_PROJECT_INDEXER_API_KEY")


from llama_index import VectorStoreIndex, SimpleDirectoryReader, ServiceContext

from llama_index.llms import OpenAI

# Necessary to use the latest OpenAI models that support function calling API
service_context = ServiceContext.from_defaults(llm=OpenAI(model="gpt-4"))
data = SimpleDirectoryReader(input_dir=".").load_data()
index = VectorStoreIndex.from_documents(data, service_context=service_context)

chat_engine = index.as_chat_engine(chat_mode="openai", verbose=True)

# Until the user types "exit", keep asking for queries
print("System ready. Type 'exit' to quit.")
while True:
    query = input("\n\n")
    query = query.strip()
    if query == "exit":
        break
    response = chat_engine.chat(query)
    print(response)

[dosubot[bot]]

🤖

Hi @buckmaxwell, good to see you again!

From your question, it seems like you're trying to use LlamaIndex to answer questions about the code in your repository. However, LlamaIndex, specifically the VectorStoreIndex class, is designed to index and query data based on vector embeddings. It's not designed to directly analyze or answer questions about the code in your repository. It works by storing and retrieving data based on vector representations, which is useful for tasks like similarity search or nearest neighbor search.

However, there are a few things you could try to improve the results:

1. You could try setting the index to be called at least once on every user message with index.as_chat_engine(..., function_call="query_engine_tool"). A system_prompt="..." kwarg may also help the bot better understand its purpose. Alternatively, you could set up the agent from scratch with a better name/description or try index.as_chat_engine(chat_mode="context"). Here is the sample code:

from llama_index import VectorStoreIndex, SimpleDirectoryReader

documents = SimpleDirectoryReader('data').load_data()
index = VectorStoreIndex.from_documents(documents)
query_engine = index.as_chat_engine(chat_mode="context")
response = query_engine.chat("What did the author do growing up?")
print(response)

response = query_engine.chat("Oh interesting, tell me more.")
print(response)

This solution was suggested in a similar issue: [Bug]: Running the chatbot customization tutorial doesn't load context data

2. To correctly index the documents, you should use the SimpleDirectoryReader class from LlamaIndex. This class can be configured to include both the filename and the document text. The filename_fn function can be used to set the metadata of each document to include the filename. Here is the corrected code:

from llama_index import SimpleDirectoryReader
filename_fn = lambda filename: {'file_name': filename}

# automatically sets the metadata of each document according to filename_fn
documents = SimpleDirectoryReader(input_files=text_list,  file_metadata=filename_fn).load_data()

print(documents[0].text)
print(documents[0].metadata)

This solution was suggested in a similar issue: [Question]: Indexes built from text don't produce good answers

3. You can enable all files to be parsed by passing use_parser = False to the constructor of GithubRepositoryReader. This was suggested in a similar issue: [feature request] GithubRepositoryReader seems to ignore all source code files

I hope this helps! Let me know if you have any other questions.

Sources

• [Bug]: Running the chatbot customization tutorial doesn't load context data
• [Question]: Indexes built from text don't produce good answers
• [feature request] GithubRepositoryReader seems to ignore all source code files
• llama_index/vector_stores/tair.py
• llama_index/tools/tool_spec/load_and_search/base.py
• llama_index/indices/vector_store/base.py

---

This response is meant to be useful and save you time. It is not meant to be a precise solution, but rather a starting point for your own research.

Help me be more useful! Please leave a 👍 if this is helpful and 👎 if it is irrelevant.

If you want to continue the conversation, start your reply with @dosu-bot.

[Disiok]

Hey @buckmaxwell, you are right that the default configuration of the project is optimized for indexing natural language text files.

You'd need to do a bit of customization to get the best performance for code. For example, it's better to use a code specific parser: https://docs.llamaindex.ai/en/stable/module_guides/loading/node_parsers/root.html#text-splitter-customization