Document API key patterns

[simonw]

Two patterns to document:

• Using datasette-secrets - ref https://github.com/datasette/datasette-secrets/issues/8
• Showing an API key field for each run if the user hasn't configured one

Plus that bit where the API key shouldn't be stored in the database for fear of it leaking by accident.

datasette-enrichments-gpt and datasette-enrichments-opencage both implement that pattern already.

New documentation should go here: https://enrichments.datasette.io/en/stable/developing.html

[simonw]

Here's how datasette-enrichments-gpt does that at the moment: https://github.com/datasette/datasette-enrichments-gpt/blob/1353a46f08f73e9b18a58115d74730fdab7427ad/datasette_enrichments_gpt/__init__.py#L32-L116

    async def get_config_form(self, datasette, db, table):
        class ConfigForm(Form):
            # Select box to pick model from gpt-3.5-turbo or gpt-4-turbo
            model = SelectField(
                "Model",
                choices=[
                    ("gpt-3.5-turbo", "gpt-3.5-turbo"),
                    ("gpt-4-turbo", "gpt-4-turbo"),
                    ("gpt-4-vision", "gpt-4-turbo vision"),
                ],
                default="gpt-3.5-turbo",
            )
            # ... more stuff like that

        def stash_api_key(form, field):
            if not (field.data or "").startswith("sk-"):
                raise ValidationError("API key must start with sk-")
            if not hasattr(datasette, "_enrichments_gpt_stashed_keys"):
                datasette._enrichments_gpt_stashed_keys = {}
            key = secrets.token_urlsafe(16)
            datasette._enrichments_gpt_stashed_keys[key] = field.data
            field.data = key

        class ConfigFormWithKey(ConfigForm):
            api_key = PasswordField(
                "API key",
                description="Your OpenAI API key",
                validators=[
                    DataRequired(message="API key is required."),
                    stash_api_key,
                ],
                render_kw={"autocomplete": "off"},
            )

        plugin_config = datasette.plugin_config("datasette-enrichments-gpt") or {}
        api_key = plugin_config.get("api_key")

        return ConfigForm if api_key else ConfigFormWithKey

Then it accesses that stashed key here:

https://github.com/datasette/datasette-enrichments-gpt/blob/1353a46f08f73e9b18a58115d74730fdab7427ad/datasette_enrichments_gpt/__init__.py#L229-L244

def resolve_api_key(datasette, config):
    plugin_config = datasette.plugin_config("datasette-enrichments-gpt") or {}
    api_key = plugin_config.get("api_key")
    if api_key:
        return api_key
    # Look for it in config
    api_key_name = config.get("api_key")
    if not api_key_name:
        raise ApiKeyError("No API key reference found in config")
    # Look it up in the stash
    if not hasattr(datasette, "_enrichments_gpt_stashed_keys"):
        raise ApiKeyError("No API key stash found")
    stashed_keys = datasette._enrichments_gpt_stashed_keys
    if api_key_name not in stashed_keys:
        raise ApiKeyError("No API key found in stash for {}".format(api_key_name))
    return stashed_keys[api_key_name]

A utility function to make this pattern easier to implement would be a good idea.

[simonw]

This is already handled in:

• 46