ki2-externai

🔒 Manage read-only external projects for AI coding assistants (Codex, Copilot, etc.)

ki2-externai is a tiny command-line tool that lets you mount external libraries or frameworks in read-only mode inside your project.
It’s especially useful when working with code-assistant tools (e.g. OpenAI Codex) that need to see your internal frameworks, but should never modify them.

🧠 Why?

When developing with AI assistants, sometimes you want the assistant to explore your internal codebases (frameworks, utils, etc.) to understand project context — but you don’t want it to modify them. ki2-externai makes this safe and transparent:

• Every external library is visible inside your project,
• but physically read-only thanks to mount --bind + remount,ro,
• and easily reversible.

Works great in Linux and WSL2.

🚀 Quick start

1. Install

# Recommended for CLI tools (isolated, global command)
pipx install ki2-externai

# Alternative (inside an existing Python environment)
pip install ki2-externai

2. Register projects once (from inside them)

cd /path/to/libA
externai register libA

This stores a name → absolute path mapping in your OS data dir (e.g. ~/.local/share/ki2/externai/registry.json). If you omit a name, the folder name is used.

3. Link externals inside your project (CLI creates config)

cd /path/to/your-project
externai add libA

If externai.json is missing, externai add creates a minimal config before appending the entry.

4. Mount everything read-only

externai mount

This creates one folder per import under .externals/ (e.g. .externals/libA). Each entry is a bind mount that is then remounted read-only.

5. Check status (or unmount)

externai status

Output example:

name                 src_exists mounted    mount_source                             ntfs-ish?
-----------------------------------------------------------------------------------------------
libA                 yes        mounted    /home/adrien/dev/libA                    no

externai unmount

⚙️ Commands

🧾 Config file (optional)

You can edit externai.json manually if you want, but you don’t have to. The CLI commands above are enough for most workflows.

Example config (package.json‑like):

{
  "version": 1,
  "externalsDir": ".externals",
  "imports": [
    "libA",
    { "name": "frameworkX", "path": "/home/adrien/dev/frameworkX" },
    { "name": "toolkit", "mount": "tk" }
  ]
}

Config semantics:

• version (number, optional): schema version. Default 1.
• externalsDir (string, optional): destination folder for mounts. Default .externals.
• imports (array, required, can be empty): list of projects to import. Each entry can be:
  • a string: the registered project name (its path is resolved from the registry),
  • or an object with at least name and optional fields:
    • path (string, optional): source path on disk (if omitted, resolved from registry),
    • mount (string, optional): name of the subfolder under externalsDir (defaults to name).

🧾 Registry details

• Run externai register [custom-name] inside the folder you want to expose. The command stores an absolute path → name mapping in your OS data dir (e.g. ~/.local/share/ki2/externai/registry.json). If you omit the name, the current folder name is used.

• externai register <name> refuses to overwrite existing entries or duplicate paths unless you confirm. Registering the exact same name/path is a no-op.

• To remove an entry, run externai unregister <name> or simply externai unregister inside the project root to drop any entry pointing to the current path.

• Use externai registry to list all registered names and their directories. You can then reference these names directly inside externai.json imports.

• Example config relying on the registry only:

  {
    "imports": ["libA", "frameworkX"]
  }
  

  All paths are injected dynamically at mount time.

🧪 Tests

• uv run pytest (ensure the tests optional dependency is installed via uv sync)

➕ Add/remove imports

• externai add <name> adds the named project to externai.json using a registered path. If the name is missing from the registry, externai explains how to register it or how to pass an explicit path (see below).
• externai add <name> <path> creates an object import for the provided path (will fail if the path does not exist).
• When externai.json is absent, externai add creates a minimal config before appending the entry.
• externai remove <name> drops the import entry with the given name.

The commands operate on the externai.json file at the project root, so you must run them from within your project (the same directory that contains the config).

💡 Tips

• Best results on WSL ext4 filesystem (not /mnt/c).
• Requires mount, umount, and optionally findmnt.
• No root install: sudo is used only for mount operations.

📦 Project structure

your-project/
├── externai.json          # config file
├── .externals/
│   ├── libA/               # bind-mounted (read-only)
│   └── frameworkX/
└── ki2_externai/
    ├── __init__.py
    ├── cli.py
    ├── config.py
    ├── operations.py
    └── utils.py

pip install ki2-externai exposes the externai console script (defined in pyproject.toml), so you can run externai mount directly.

🪪 License

MIT License.