explore-cli 0.3.2

Interactively explore a codebase with an LLM

explore

Interactively explore a codebase with an LLM

explore is a script to interactively explore a codebase by chatting with an LLM. It uses retrieval-augmented generation via chromadb to provide the LLM with relevant source code from the codebase.

explore uses OpenAI models by default, so you'll need an OpenAI API key.

Installation

explore is available on PyPI. I recommend installing it with pipx:

pipx install explore-cli
export OPENAI_API_KEY=<your OpenAI API key>
explore <directory>

Alternatively, you can clone this repository and run the script with poetry:

poetry install
poetry build
export OPENAI_API_KEY=<your OpenAI API key>
poetry run explore <directory>

Usage

usage: explore [-h] [-l LLM] [-m MODEL] directory

Interactively explore a codebase with an LLM.

positional arguments:
  directory             The directory to index and explore.

options:
  -h, --help            show this help message and exit
  -l LLM, --llm LLM     The LLM backend, one of openai or ollama. Default: openai
  -m MODEL, --model MODEL
                        The LLM model to use. Default: gpt-4o-mini for openai or mistral-nemo:latest for ollama

How it works

1. The codebase is indexed into a local Chroma store. Each file is split into chunks using language-specific separators.
2. Documents relevant to the query are collected using multiple retrieval strategies:
   • Primary retrieval is done through vector similarity using the indexed embeddings.
   • A multi-query retriever issues multiple variations of the query to increase the diversity and relevance of retrieved documents.
   • Additionally, a history-aware retriever reformulates the user query, considering the conversation history to better capture context.
3. Retrieved documents are deduplicated, concatenated, and added as context to the LLM, which generates an answer to the user's question. Answers include specific references to the files and code pertinent to the query.