Code Assistant

A tool for code analysis and generation using large language models.

Installation

pip install code-assistant

Usage

The package provides a command-line interface with several subcommands. It's generally easier to add secrets to a .env file in the project root directory than to pass them as arguments.

Codebases are stored in a MongoDB database, although in future more storage options will be provided.

The following environment variables can be set:

• MONGODB_URL
• OPENAI_API_KEY

Extract Code from GitHub

# Extract code from a GitHub repository
code-assistant extract github \
    --repo-url="https://github.com/username/repo" \
    --github-token="your_token"  # Optional, for private repos

Generate Embeddings

# Generate embeddings for code units
code-assistant embed generate \
    --input-path="code_units.json" \
    --model-name="jinaai/jina-embeddings-v3" \
    --codebase="your-codebase-name"

# Compare a query against embedded code
code-assistant embed compare \
    --query="How do I handle errors?" \
    --model-name="jinaai/jina-embeddings-v3" \
    --codebase="your-codebase-name"

Generate Training Data

# Generate prompt-code pairs using OpenAI
code-assistant generate prompts \
    --output-path="prompt_code_pairs.json" \
    --num-rows=100

Evaluate Retrieval

# Evaluate retrieval performance
code-assistant evaluate retrieval \
    --test-data-path="test_prompt_code_pairs.json" \
    --codebase-path="embedded_code_units.json" \
    --output-path="evaluation_results.json"

RAG Prompt (End-to-End)

Basic usage

code-assistant rag prompt \
    --query="How do I handle errors in this codebase?" \
    --codebase="your-codebase-name"

Advanced usage with all options