LLMCheatsheets.jl

LLMCheatsheets.jl

LLMCheatsheets.jl is a Julia package that makes it easy and instant to teach AI models about new packages and repositories by creating cheatsheets from GitHub repositories. This tool bridges the gap between human-readable documentation and AI-friendly knowledge representation, allowing for seamless integration with language models and AI assistants.

By default, we take a subset of the folders and files in the provided repository and summarize them using an LLM into a single cheatsheet.

Features

• Instant cheatsheet generation from GitHub repositories.
• AI-friendly knowledge representation by summarizing code and documentation (or just collate all the raw files into a single string).
• Easy integration with language models and AI assistants (just copy the cheatsheet into your prompt).
• Support for any package and easier to start than Retrieval Augmented Generation (RAG).

Installation

To install LLMCheatsheets.jl, use the Julia package manager and the repo URL (it's not registered yet):

using Pkg
Pkg.add(url = "https://github.com/svilupp/LLMCheatsheets.jl")

Tip

If you encounter rate limits when accessing the GitHub API, you can set up a personal access token and set it as an environment variable GITHUB_API_KEY to increase your request limit to 5000 per hour.

Quick Start

Here's a basic example of how to use LLMCheatsheets.jl to create a cheatsheet for a GitHub repository.

using LLMCheatsheets

repo = GitHubRepo("https://github.com/svilupp/PromptingTools.jl")
create_cheatsheet(repo; save_path = true);

With save_path = true, the cheatsheet will be saved to folder llm-cheatsheets in the current working directory.

What happens behind the scenes:

1. Scanning the Repository: The repository is scanned to find all relevant files that match repo.paths and repo.file_types.
2. Summarizing Files: Each file is summarized using an LLM.
3. Generating Cheatsheet: The summaries are combined to generate a comprehensive cheatsheet.

For a low-level interface to generate the files individually and process them yourself, see examples/create_for_promptingtools.jl.

Sometimes you might want to just download the files without summarizing them. You can do that with collect function.

files_str = collect(repo)

files_str will be a string with all scanned files concatenated together. To use it in ChatGPT or Claude.ai, use clipboard functionality to copy it to clipboard - files_str|>clipboard.

By default, the files scanned and downloaded are repo.paths and repo.file_types, respectively.

Advanced Usage

Customizing scanned paths and file types

By default, repo.paths includes ["src", "docs/src", "README.md"], and repo.file_types includes [".jl", ".md"]. You can customize these when creating the GitHubRepo object:

Eg, adding a folder examples and .txt files to customize what we will summarize:

repo = GitHubRepo("https://github.com/username/repository"; paths = ["examples", "README.md"], file_types = [".jl", ".md", ".txt"])

Using a different LLM

You can use a different LLM by passing the model argument to the functions.

create_cheatsheet(repo; save_path = true, model = "gpt4om")

Adding Special Instructions

You can provide special instructions to guide the AI in generating the cheatsheet:

create_cheatsheet(repo; special_instructions = "Focus on the data structures and their interactions.")

Using PromptingTools.jl to Ask Questions

You can simply export also ai* functions from PromptingTools.jl to use them with LLMCheatsheets.jl.

using LLMCheatsheets
# Re-export aigenerate, pprint from PromptingTools
using LLMCheatsheets: aigenerate, pprint
# Or import PromptingTools directly
# using PromptingTools

repo = GitHubRepo("https://github.com/svilupp/PromptingTools.jl"; paths = ["docs/src", "README.md"])
files_str = collect(repo)

msg = aigenerate("Read through these files: $(files_str)nnAnswer the question: What is the function for creating prompt templates?")
pprint(msg)

The function for creating prompt templates in the `PromptingTools.jl` package is `create_template`. This function allows you to define a prompt with placeholders and save it for later use. The syntax is:

```julia
create_template(; user=<user prompt>,
system=<system prompt>, load_as=<template name>)
```

This function generates a vector of messages, which you can use directly in the `ai*` functions. If you provide the `load_as` argument, it will also register the template in the template store,
allowing you to access it later using its name.

Frequently Asked Questions

I am getting rate-limited by LLM providers

If you are getting rate-limited by LLM providers, you can decrease the number of concurrent summarization tasks in create_cheatsheet by setting a lower number like ntasks=5 or ntasks=2 (depends on your API tier).

I am getting rate-limited by GitHub API

Set up a personal access token and set it as ENV["GITHUB_API_KEY"]. It will be automatically loaded into a variable LLMCheatsheets.GITHUB_API_KEY.

How do I set up a personal access token for GitHub API?

You can set up a personal access token for GitHub API by following these steps:

1. Go to your GitHub settings.
2. Click on "Personal access tokens".
3. Click on "Generate new token".

Then you can set it as ENV["GITHUB_API_KEY"] or LLMCheatsheets.GITHUB_API_KEY.

Documentation

For more information about using LLMCheatsheets.jl, please refer to the documentation.

Issues and Support

If you encounter any issues or have questions, please open an issue on the GitHub repository.