Docstring AI — Python Docstrings, Offline + AI

autoDocstring-compatible skeletons — plus AI that actually fills them in.

Docstring AI generates Python docstrings in VS Code two ways:

1. Skeleton mode — instant, fully offline. Type """ under a function and get a correctly structured skeleton built from the signature and body: parameters, types, defaults, returns/yields, raised exceptions.
2. AI mode — optional, bring your own API key. One command sends the function to your model (Anthropic or any OpenAI-compatible endpoint) and inserts a docstring with real descriptions, formatted in your chosen style.

If you used autoDocstring (unmaintained for years), this is the drop-in workflow you already know — same trigger, same styles, same skeleton behavior — under active maintenance, with AI as a strictly opt-in extra.

Before / after

def fetch_rows(query: str, limit: int = 100, *, retries: int = 3) -> list[dict]:
    if not query:
        raise ValueError("empty query")
    return _run(query, limit, retries)

Type """ on the line below the signature, accept Generate Docstring, and skeleton mode inserts (Google style):

def fetch_rows(query: str, limit: int = 100, *, retries: int = 3) -> list[dict]:
    """_summary_

    Args:
        query (str): _description_
        limit (int, optional): _description_. Defaults to 100.
        retries (int, optional): _description_. Defaults to 3.

    Returns:
        list[dict]: _description_

    Raises:
        ValueError: _description_
    """
    if not query:
        raise ValueError("empty query")
    return _run(query, limit, retries)

Every _summary_ / _description_ / _type_ is a snippet tabstop — Tab through and fill them in. Or run Docstring AI: Generate with AI and the placeholders arrive already written.

Skeleton mode vs AI mode

Both modes support Google, NumPy, and Sphinx formats, functions, async functions, methods, and classes (attributes are collected from __init__ assignments and class-level annotations).

Commands

Settings

You rarely need to edit the three AI settings by hand: run Docstring AI: Choose AI Provider & Model and pick a preset — it writes them for you (globally) and offers to store your API key if none is set yet.

Privacy — where your code goes

• Skeleton mode never touches the network. Parsing and generation run entirely inside the extension host. You can use it on an air-gapped machine.
• Code leaves your machine only when you explicitly run "Docstring AI: Generate with AI". That command sends the selected function/class source (plus the enclosing class name and its docstring, for context) directly to the provider you configured — nothing else, nowhere else.
• There is no middleman server and no telemetry. Requests go straight from your editor to api.anthropic.com or the base URL you set. Nothing is logged, collected, or phoned home by the extension.
• Your API key is stored only in VS Code Secret Storage (your OS keychain). It is never written to settings, files, or workspace state.

Roadmap

• Paid tier (planned): batch mode (document a whole file/package in one pass), docstring style linting, and CI coverage checks for undocumented public APIs. The skeleton generator stays free, offline, and unlimited.

For autoDocstring users

Behavior compatibility is a design goal: the """ trigger, the _summary_/_description_/_type_ placeholders, Defaults to … phrasing, and the Google/NumPy/Sphinx section layouts all match the workflow you're used to, so your muscle memory (and your team's docstring conventions) carry over. This project shares no code or text with autoDocstring — it is an independent, from-scratch implementation.

License

MIT