ctxs.ai

What is ctxs.ai

ctxs.ai is an open-source, community-curated directory of contexts for use with LLMs.

Users can contribute context data as markdown files via GitHub pull requests.

Rationale

• LLMs are trained with huge datasets, constrained by a training cut off date
• Providing more precise context about a project, libraries used, available tools and frameworks improves model performance
• IDEs like Cursor are starting develop their own indexing system but this makes it hard to share what’s working
• Serve as an open-source “deterministic checkpoint” that can be used for evals

Use Cases

Generally this exists to share context windows and prompts, see what works for others and learn from it.

Specifically for coding use cases:

• Help LLMs work more efficiently in a specific framework
• Define coding style, conventions and formatting
• Provide library documentation in a format suited for LLMs

The Repository

The repo contains a namespaces tree of markdown files which can be contributed via PR. Each markdown file comes with certain frontmatter (last updated, description, whether it is ai generated, a link to how it was generated)

Currently you are invited to contribute under your personal namespace (GitHub username). I would love to see contributions to teach LLMs how to work with specific frameworks, translate code between languages or libraries,

If you're unsure how to contribute feel free to open an issue with your contribution.

How to contribute

ctxs.ai is meant as an open registry for context windows, for now you can add contexts under

./contexts/{your github username}/{context-slug}.md

To add a file use the following template

---
title: 
description: 
provenance: 
type: "prompt" or "context"
---
the actual context prompt

provenance is meant to provide some background about how the context window was created (hand written, generated, edited, etc.)

Interface

For now ctxs.ai redirects to this repo. Eventually it will serve as a CDN. Think of unpkg for prompts and context windows.

Some ideas:

• A command line script to search and download contexts
• A Model Context Protocol server
• A web UI to browse and download

For now the README will contain a list of all entries.

Inspirations

• Jim Lowin The Qualified Self
• llms.txt and llmstxt.directory
• PatrickJS' awesome-cursorrules
• Daniel Rosehill Prompt Library (read the README, it's great)
• shadcn CLI experience

Like this project?

? Star the repo and contribute! ?

Available Contexts

Browse via GitHub for now (type t now). Will fix soon :-)

All contexts can also be retrieved via curl, e.g.

curl https://ctxs.ai/gh/martinklepsch/babashka