Intent-driven generation

udiot create (alias c) asks a model what you described, then writes it with the same generators as udiot generate.

npx --no-install udiot create --help
npx --no-install udiot create "an order service that lists open orders" --dry-run

Setup

Enable AI in udiot.config.json and provide a key:

{ "framework": { "ai": { "enabled": true, "model": "gpt-4o-mini" } } }

export OPENAI_API_KEY=...
# Optional: any OpenAI-compatible endpoint
export OPENAI_BASE_URL=https://llm.internal.example/v1

Requests go through the framework's LLM client, so the openai package (an optional peer) must be installed. framework.ai.temperature and framework.ai.contextWindow (used as the response token limit) apply.

What happens

1. The description is sent to the model, which returns the type (component, service, controller, event or handler), a name, and props, methods, endpoints or event fields.
2. The matching generator renders the files. For components, services and controllers the model then writes each method body; a body it cannot write stays a TODO stub.
3. The generated files are sent back to the model for review, and its findings are reported. Pages, features and modules are not generated.

Options are those of generate: --dry-run, --yes, --force, --with-tests, --description, --event-name, --event-category, --event-version, --event-docs and --for-event. Writing asks for confirmation unless --yes or --dry-run is given.

--dry-run previews the files; the model is still called, so it still costs tokens and still sends your description to the provider.

Treat the output as scaffolding: model-written code can be wrong. Review imports, authorization, validation and persistence, compile it, and test it. Do not put credentials or private data in descriptions.