GitHub - laiso/askrepo: Source code reading with LLM.

3 min read Original article ↗

askrepo - Source code reading with LLM

This program reads the content of Git-managed text files in a specified directory, sends it to the Google Gemini API, and provides answers to questions based on the specified prompt. It supports streaming responses for real-time feedback.

Usage

❯ askrepo --help
Usage: askrepo [options] [base_path]

Arguments:
  base_path                  Directory containing source code files

Options:
  -p, --prompt <TEXT>        Question to ask about the source code [default: "Explain the code in the files provided"]
  -m, --model <TEXT>         Google AI model to use [default: "gemini-1.5-flash"]
  -a, --api-key <TEXT>       Google API key
  -u, --base-url <TEXT>      API endpoint URL [default: "https://generativelanguage.googleapis.com/v1beta/models/"]
  --stream                   Enable/disable streaming mode [default: true]
  -v, --verbose              Enable verbose output
  -h, --help                 Show help

Installation

Using Deno

Run directly without installation

# Run directly using Deno
export GOOGLE_API_KEY="YOUR_API_KEY"
deno run -A jsr:@laiso/askrepo --prompt "What is this code doing?"  ../your-repo/src

Install globally

# Install globally using Deno
deno install -A --global jsr:@laiso/askrepo

# Make sure $HOME/.deno/bin is in your PATH
export PATH="$HOME/.deno/bin:$PATH"

# Then run the command
export GOOGLE_API_KEY="YOUR_API_KEY"
askrepo --prompt "What is the purpose of this code?" ../your-repo/src

Gemini API Key

A Gemini API key is required to run this program. You can get it from:

https://aistudio.google.com/app/apikey

Examples

Using Google Gemini API (default)

export GOOGLE_API_KEY="YOUR_API_KEY"
askrepo --prompt "What is the purpose of this code?" ../your-repo/src

# Using short options
askrepo -p "What is the purpose of this code?" -m "gemini-2.0-flash" ../your-repo/src

Using OpenAI API

export OPENAI_API_KEY="YOUR_OPENAI_API_KEY"
askrepo --prompt "What is the purpose of this code?" \
  --model "o3-mini" \
  --base-url "https://api.openai.com/v1/chat/completions" \
  ../your-repo/src

# Using short options
askrepo -p "What is the purpose of this code?" \
  -m "o3-mini" \
  -u "https://api.openai.com/v1/chat/completions" \
  ../your-repo/src

Development

# Run the project in development mode
deno run -A mod.ts --prompt "Find bugs in this code" ./src

Run Tests

Implementation Details

Git-managed File Tracking

The tool uses the .gitignore parser to determine which files to include:

  • Scans directories recursively while respecting .gitignore rules at each level
  • Skips binary files, hidden files, and node_modules directories
  • Caches .gitignore rules to optimize performance

Binary File Detection

Files are identified as binary through two complementary methods:

  1. Extension-based detection: Checks if the file extension matches known binary formats (jpg, png, etc.)
  2. Content-based detection:
    • Looks for null bytes in the first 1024 bytes
    • Identifies binary file signatures (magic numbers) for common formats like JPEG, PNG, and GIF

File Content Processing

  • Reads text content from all non-binary tracked files
  • Double-escapes content to ensure proper JSON formatting
  • Combines file content into a tab-separated format with filename references

LLM Integration

  • Constructs a structured prompt that includes:
    • File paths and their contents
    • The user's question
    • Instructions for the model to reference specific files in its response
  • Supports both Google's Generative AI models (Gemini) and OpenAI-compatible APIs

Streaming Response Handling

  • Uses Deno's native fetch API with streaming support
  • Processes chunk-based responses in real-time
  • Parses SSE (Server-Sent Events) data format from the API
  • Handles both streamed and non-streamed responses

Command Line Interface

  • Built on Deno's standard library for parsing arguments
  • Provides sensible defaults for all options
  • Auto-detects API keys from environment variables
  • Supports both standard and shorthand options
  • Returns helpful error messages for invalid inputs

Features

  • Streaming support for real-time AI responses
  • Flexible API endpoint configuration
  • Improved response parsing for different formats
  • Support for both streaming and non-streaming modes
  • Default to the latest Gemini model (gemini-2.0-flash), but can be configured to use other models
  • Support for OpenAI Compatible API