The Smart Form Filler fills in form fields based on unstructured user data, e.g. email content.
Add a call to provideFormFiller()
in your application config to use the form filler.
You need to specify a backend to use the form filler. Currently, there are the following backends:
- The OpenAI backend (
withOpenAIBackend()
) talks to a OpenAI-compatible implementation such as OpenAI, Azure OpenAI, MistralAI, Groq, and others. It requires Internet access. - The OpenAI tool backend (
withOpenAIToolBackend()
) behaves as the OpenAI backend, but makes use of tools (previously function calling). - The experimental WebLLM backend (
withWebLLMBackend()
) uses WebLLM to execute a large language model locally. The initial download takes a relatively long time, but model inference is offline-capable from then on. The model download must be repeated for each origin. - The experimental Prompt API backend (
withPromptAPIBackend()
) uses Chrome’s experimental Prompt API to communicate with a local LLM. The model, shared across origins, cannot be specified but is fully offline-capable once downloaded.
The demo project is configured to use OpenAI by default. You can also use another OpenAI-compatible provider (e.g., Groq).
Before running the demo, you will need to set your OpenAI key by using the environment variable OPENAI_API_KEY
.
Use the same variable to set your API key for any other OpenAI-compatible provider.
Run npm start
.
This project was heavily inspired by Steve Sanderson’s Smart Components for Blazor.
BSD-3-Clause.
This is a technical showcase, not an official Thinktecture product.