DEV Community

Lakshmi Sravya Vedantham
Lakshmi Sravya Vedantham

Posted on

I built a CLI that gives any AI instant context about your project

#python #ai #claude #devtools

Every time I start a new AI session, I spend the first few minutes explaining the same things:

  • "This is a FastAPI project"
  • "We use SQLAlchemy for the ORM"
  • "The main entry point is src/api/main.py"
  • "Recent work has been on the auth module"

It's tedious. And AI tools like Claude Code, ChatGPT, and Gemini start cold every session.

So I built ctx.

pip install ctx

ctx save myproject          # scan project, save as context pack
ctx inject myproject        # paste into any AI chat instantly
ctx inject myproject --target claude  # write CLAUDE.md for Claude Code
Enter fullscreen mode Exit fullscreen mode

What it does

ctx save scans your project and builds a context pack automatically:

  • Stack detection — finds pyproject.toml, package.json, Cargo.toml, go.mod, Gemfile, etc.
  • Structure map — directory tree of your src/, tests/, api/ folders
  • Git log — last 10 commits so the AI understands what you've been working on
  • README summary — first few lines as project context
  • Your notes — add anything extra on top

The result is a clean Markdown file that any AI can parse immediately.

Inject anywhere

ctx inject myproject puts the context pack where you need it:

ctx inject myproject                      # → clipboard (paste into ChatGPT, Gemini, etc.)
ctx inject myproject --target claude      # → writes CLAUDE.md in current directory
ctx inject myproject --target chatgpt     # → clipboard, formatted as a system prompt
Enter fullscreen mode Exit fullscreen mode

Claude Code reads CLAUDE.md automatically when you open a project. No paste needed.

For ChatGPT, Gemini, or anything else — one paste at the start of the session and you're fully loaded.

What a pack looks like 

# myproject

## Stack
- Python
- Detected from: pyproject.toml

## Structure
src/
  api/
  models/
tests/

## Recent commits
- feat: add user auth
- fix: resolve migration conflict
- refactor: extract service layer

## README
MyProject is a FastAPI app for managing...

## Notes
Main entry: src/api/main.py
Auth lives in src/auth/ — JWT-based, no sessions
Enter fullscreen mode Exit fullscreen mode

Global vs local packs 

ctx save myproject --scope global   # ~/.ctx/packs/myproject.md (default, any directory)
ctx save myproject --scope local    # .ctx/myproject.md (project-specific, git-committable)
Enter fullscreen mode Exit fullscreen mode

Local packs take priority. Commit .ctx/ to your repo and your whole team gets the same context.

The full command set 

ctx save myproject          # scan + save
ctx list                    # show all packs
ctx show myproject          # print pack to terminal
ctx inject myproject        # inject (clipboard by default)
ctx edit myproject          # open in $EDITOR
ctx delete myproject        # remove pack
Enter fullscreen mode Exit fullscreen mode

Why this matters

"Context engineering" is the new prompt engineering. The models are good. What holds them back is not having enough context about your project — your conventions, your current work, your architecture decisions.

ctx is a local, zero-dependency way to fix that. No account. No sync service. Just Markdown files you control.

Try it 

pip install ctx

# In any project
ctx save myproject --notes "Add anything you want the AI to know"
ctx inject myproject --target claude
Enter fullscreen mode Exit fullscreen mode

Source: github.com/LakshmiSravyaVedantham/ctx

What's your biggest friction starting an AI session on an existing codebase? Drop it in the comments.

Top comments (0)