/* - this version collects data from three ways: a free-form paragraph, a JSON text area, or a table (mimicking a form) - it's really tempting to create a form that lets people "write anything" in a text box and have the data properly formatted by the AI, it's still kind of an accessibility nightmare. - I can see this appearing everywhere, maybe for the novelty, but current forms with labels and ARIA etc. will still reign supreme for data-heavy bureaucratic data collection tasks - for casual inputs though like guest books this could be a simpler alternative - for "real data applications" like collecting info for shipping and CC, still use a "real" form - for "real applications" this can still be used to catch errors quickly, like misspellings (which happen a LOT) - using a really fast/cheap model like llama3.1 8b (or even smaller) is a good use case, but it should support structured outputs */ import { Hono } from "https://deno.land/x/hono@v3.11.7/mod.ts"; const app = new Hono(); const module = ``; const html = ` Squishy Form Demo

Squishy Form Experiment

This is an experiment of a form that can parse between free-form text, JSON, and table data, and output a structured JSON object. It uses Groq Llama 3.1 8b as a text parser, and uses nanostore to manage state and immer to manage undo/redo.

Input Text System Text

This is the system prompt that will be used to guide the parser. It's a comma-separated list of keys that will be used to create the output JSON.

A labspace.ai experiment

${module} `; app.get("/", (c) => c.html(html)); export default (typeof Deno !== "undefined" && Deno.env.get("valtown")) ? app.fetch : app;

squishyform

Version 3